Phase separation of Arabidopsis EMB1579 controls transcription, mRNA splicing, and development

Tight regulation of gene transcription and mRNA splicing is essential for plant growth and development. Here we demonstrate that a plant-specific protein, EMBRYO DEFECTIVE 1579 (EMB1579), controls multiple growth and developmental processes in Arabidopsis. We demonstrate that EMB1579 forms liquid-like condensates both in vitro and in vivo, and the formation of normal-sized EMB1579 condensates is crucial for its cellular functions. We found that some chromosomal and RNA-related proteins interact with EMB1579 compartments, and loss of function of EMB1579 affects global gene transcription and mRNA splicing. Using floral transition as a physiological process, we demonstrate that EMB1579 is involved in FLOWERING LOCUS C (FLC)-mediated repression of flowering. Interestingly, we found that EMB1579 physically interacts with a homologue of Drosophila nucleosome remodeling factor 55-kDa (p55) called MULTIPLE SUPPRESSOR OF IRA 4 (MSI4), which has been implicated in repressing the expression of FLC by forming a complex with DNA Damage Binding Protein 1 (DDB1) and Cullin 4 (CUL4). This complex, named CUL4-DDB1MSI4, physically associates with a CURLY LEAF (CLF)-containing Polycomb Repressive Complex 2 (CLF-PRC2). We further demonstrate that EMB1579 interacts with CUL4 and DDB1, and EMB1579 condensates can recruit and condense MSI4 and DDB1. Furthermore, emb1579 phenocopies msi4 in terms of the level of H3K27 trimethylation on FLC. This allows us to propose that EMB1579 condensates recruit and condense CUL4-DDB1MSI4 complex, which facilitates the interaction of CUL4-DDB1MSI4 with CLF-PRC2 and promotes the role of CLF-PRC2 in establishing and/or maintaining the level of H3K27 trimethylation on FLC. Thus, we report a new mechanism for regulating plant gene transcription, mRNA splicing, and growth and development.


Introduction
Plant growth and development are tightly regulated in response to many endogenous and environmental signals by genetic and cellular programs that determine plant form. As sessile organisms, plants need to efficiently organize various cellular events to cope with the ever-changing (SPA) complex [39] to control the abundance of CONSTANS protein [40,41], both CUL4 and MSI4 are required to maintain the level of H3K27me3 on FLC chromatin [36]. It was also demonstrated that CUL4-DDB1 MSI4 physically associates with a CLF-PRC2 complex [36]. Therefore, the emerging scenario is that MSI4 forms the CUL4-DDB1 MSI4 complex, which physically interacts with CLF-PRC2 to establish and/or maintain the level of H3K27me3 on the FLC locus to control the expression of FLC [36].
Here we report that the plant-specific protein EMBRYO DEFECTIVE 1579 (EMB1579), which was uncovered during the systematic identification of genes required for normal embryo development in Arabidopsis [42], is able to undergo LLPS in vitro and in vivo. EMB1579 condensates exhibit liquid-like properties and turn over extremely rapidly within the nucleus. We found that many nuclear proteins crucial for chromosomal function and RNA biology interact with EMB1579 condensates and some of them colocalize with EMB1579 condensates in the nucleus. Loss of function of EMB1579 alters global gene transcription and mRNA splicing, which provides an explanation for why emb1579 mutants exhibit pleiotropic developmental defects. Using floral transition as the representative physiological process, we demonstrate that EMB1579 is involved in regulating the level of H3K27me3 on FLC and the expression of FLC. EMB1579 likely controls the function of CLF-PRC2 via direct interaction with CUL4-DDB1 MSI4 . We propose that EMB1579 condensates condense important biomolecules in the Arabidopsis nucleus to regulate their functions in controlling key nuclear events, such as gene transcription and mRNA splicing. Our study thus reveals a new mechanism for the regulation of plant growth and development through LLPS of EMB1579.

Loss of function of EMB1579 induces pleiotropic growth and developmental defects in Arabidopsis
EMB1579 initially caught our attention because it encodes a protein that is homologous to the tobacco protein MAP190, which interacts with both actin filaments and microtubules [43]. It was also proposed to be involved in nuclear calcium signaling during the salt response, as it contains an EF-hand motif [44]. To gain insights into the developmental functions of EMB1579, we examined its tissue expression pattern and found that it is widely expressed, especially in highly proliferative tissues, including embryos, the root meristematic region, the basal region of shoots, and the base of the cauline-leaf branch (S1 Fig). To examine the function of EMB1579, we characterized two T-DNA insertion lines that were shown to be knockout alleles ( Fig 1A). We observed embryonic developmental defects at different stages, resulting from distorted cell division and cell expansion in emb1579 embryos compared to wild type (WT) (Fig 1B). Specifically, we found that the division plane of cells was mispositioned and cells were swollen in emb1579 mutants ( Fig  1B). Consequently, the development of seeds and seedlings was defective in emb1579 mutants (Fig 1C-1E). Consistent with the distorted cell phenotypes in embryos, we found that the cell files were altered in the roots of emb1579 mutants ( Fig 1F). Notably, we found that the number of meristematic cells was decreased significantly in emb1579 mutants compared to WT (Fig 1G). We also found that the floral transition was delayed in emb1579 mutant plants (Fig 1H and 1I). Thus, our study suggests that EMB1579 is crucial for Arabidopsis growth and development.

EMB1579 forms highly dynamic liquid-like condensates in the nucleus, and phase separates in vitro
We next generated a functional EMB1579-tandem copies of enhanced green fluorescent protein (TGFP) fusion protein (S2 Fig). We found that it localizes to the nucleus (Fig 2A and 2B) Two T-DNA insertion lines, CS16026 and Salk_007142, were designated as emb1579-1 and emb1579-3, respectively. The positions of the T-DNA insertions are indicated by inverted triangles. Three independent pairs of primers were used to identify truncated EMB1579 transcripts in emb1579-1 and emb1579-3. The positions of the primers are indicated under the gene. The expression of EMB1579 in WT and emb1579 mutants was also confirmed by qRT-PCR analysis with primer pairs EMB1579-qRT-F1/EMB1579-qRT-R2 (S4 Table). Data are presented as mean ± s.e.m, n = 3. Numerical data underlying the graph are available in S1 Data. The original pictures are available in S1 Raw Images. (B) Micrographs of embryos at different stages. Embryos at the 8-cell stage, 16-cell stage, and globular stage were revealed by whole-mount clearing methods, and embryos at the triangular stage, heart stage, and torpedo stage were revealed by staining with PI as described previously [45]. In emb1579 mutants, the swollen cells are outlined with green lines, and white arrowheads indicate the formation of abnormal cell plates. Bars = 50 μm. (C) Images of Arabidopsis seeds. White arrowheads indicate dry wrinkled seeds. Bar = 1 mm. (D) Images of Arabidopsis seedlings. Bar = 0.5 cm. (E) Quantification of primary root length of 7-day-old seedlings in WT, emb1579-1, and emb1579-3. Data are presented as and undergoes dynamic changes during the cell cycle (Fig 2C, S1 Movie). Specifically, the intense nuclear signal of EMB1579-TGFP disappeared with the breakdown of the nuclear membrane. The nuclear localization of EMB1579 is consistent with the presence of a nuclear localization signal (NLS) in the protein sequence [46]. We found that EMB1579 forms compartments within the nucleus of Arabidopsis root cells (Fig 2D) and the size of the compartments differs between proliferative and differentiated Arabidopsis root cells (Fig 2D and 2E). EMB1579 compartments are distinct from Cajal bodies and HYL1 bodies (S3 Fig). Strikingly, we found that EMB1579 compartments are extremely dynamic, as assayed with fluorescence recovery after photobleaching (FRAP) experiments ( Fig 2F and 2G, S3 Movie), and EMB1579 compartments are able to undergo fusion ( Fig 2H, S2 Movie). These data suggest that EMB1579 compartments exhibit liquid-like properties in the nucleus.
Next, we wondered whether EMB1579 on its own can phase separate in vitro. We initially analyzed the amino acid sequence of EMB1579 using different phase-separation predictors reported previously [47,48] and found that it contains IDRs that occupy most of the protein (Fig 2I). In addition, we found that EMB1579 contains few hydrophobic residues (Fig 2I), which may lead to weak hydrophobic interactions that will facilitate LLPS to drive the formation of compartments as shown previously [49,50]. Indeed, we found that full-length recombinant EMB1579 protein ( Fig 2J) forms highly spherical liquid-like condensates ( Fig 2K). Strikingly, the condensates are able to fuse together ( Fig 2K, S4 Movie). In particular, aqueous solutions of EMB1579 spontaneously form condensates at high protein concentrations in a dose-dependent manner, even when the salt concentration is comparatively high (Fig 2L). EMB1579 is able to undergo LLPS at 25 nM (Fig 2L), which suggests that the critical concentration for EMB1579 to undergo LLPS is very low. Interestingly, we estimated that the concentration of EMB1579 in the nucleus is about 33 nM (S4 Fig), which suggests that EMB1579 on its own can undergo LLPS in the nucleus. FRAP experiments showed that EMB1579 within condensates dynamically exchanged with EMB1579 in the surrounding environment ( Fig  2M), with an average half-time of recovery of 2.61 seconds (Fig 2N). We next performed the experiment described as "half-bleach" [18,51,52] to test whether the components within a liquid condensate undergo constant mixing. We bleached roughly half of an EMB1579 condensate and monitored its fluorescence over time. We found that EMB1579 rapidly redistributed from the unbleached area to the bleached area ( Fig 2O, S5 Movie), which suggests that EMB1579 molecules can diffuse freely within condensates and they are in a liquid state. These data together suggest that EMB1579 undergoes LLPS in vitro.

Many nuclear proteins bind to EMB1579 condensates, and loss of function of EMB1579 affects global transcription and mRNA splicing in Arabidopsis
In order to understand how EMB1579 performs its cellular functions, we next asked which proteins enter EMB1579 condensates. Using the conditions under which EMB1579 phase separates in vitro, we incubated EMB1579 with an Arabidopsis root total extract and performed pull-down experiments followed by mass spectrometry. We found that many nuclear proteins mean ± s.e.m. ��� P < 0.001 by Student t test. Numerical data underlying this panel are available in S1 Data. (F) Images of Arabidopsis roots revealed by staining with PI. White arrowheads indicate formation of abnormal cell plates. Bar = 25 μm. (G) Quantification of meristem cell number of 3-day-old seedling roots in WT, emb1579-1, and emb1579-3. Data are presented as mean ± s.e.m. ��� P < 0.001 by Student t test. Numerical data underlying this panel are available in S1 Data. (H) Images of 6-week-old Arabidopsis plants. Bar

PLOS BIOLOGY
Phase separation of EMB1579 controls transcription and mRNA splicing in plants White arrows indicate the time points that EMB1579 starts to disappear from the nucleus, whereas black arrows indicate the time points that EMB1579 starts to appear in the nucleus. Bar = 5 μm. (D) EMB1579 forms bodies within the nucleus of Arabidopsis root cells from the meristem zone and the elongation zone. Bar = 5 μm. (E) Histograms of size distribution of EMB1579 bodies in the nucleus from cells in the root meristem zone and the elongation zone. More than 200 bodies were measured in at least 12 nuclei. The average values are presented as mean ± s.e.m. Numerical data underlying this panel are available in S1 Data. (F) Images of an Arabidopsis nucleus before and after bleaching. The photobleached region is boxed. Bar = 5 μm. (G) Plot of fluorescence intensity before and after photobleaching. The blue curve represents the average value of fluorescence intensity of 18 bodies from 10 seedlings. All data are presented as mean ± s.e.m. Numerical data underlying this panel are available in S1 Data. (H) Images of an Arabidopsis nucleus showing the fusion of two EMB1579 bodies. Boxed region indicates two bodies that undergo fusion. Bar = 5 μm. (I) Analysis of the intrinsic disorder tendency and hydropathicity of the full-length EMB1579 protein. The intrinsic disorder (red line) was analyzed with IUPred2A. The scores are assigned between 0 and 1, and a score above 0.5 indicates disorder. Three long stretches of disordered regions are shaded in light blue. The hydropathicity score (blue line) was determined with ExPASy, which used the Kyte-Doolittle scale of amino acid hydropathicity with a sliding window size of 21. The scores were normalized from 0 to 1. A score above 0.5 indicates hydrophobicity. Numerical data underlying this panel are available in S1 Data. (J) Purified crucial for chromosomal function and RNA biology are enriched in the pull-down fraction ( Fig 3A, S1 Table), which suggests that EMB1579 might function by regulating the activities of those proteins. Next, we determined whether the transcriptional profile of the genes encoding those proteins was altered in emb1579 mutants. We performed RNA sequencing (RNA-seq) analysis on 7-day-old Arabidopsis seedlings and found that loss of function of EMB1579 caused differential expression of many genes involved in essential processes, including chromosome organization, DNA and RNA binding, transcription, histone binding and modification, cytoskeletal organization, and so on ( Fig 3B, S2 Table). The differential expression of 17 previously characterized genes was validated by quantitative reverse transcription PCR (qRT-PCR) analysis ( Fig 3C). To determine whether pre-mRNA splicing events were altered in emb1579 mutants, we searched our RNA-seq data for splicing defects and found that there was a total of 2,507 abnormal splicing events in emb1579 mutants, which could be divided into five different patterns (Fig 3D, S3 Table). Among them, skipped exons and retained introns were most common in emb1579 mutants ( Fig 3D). We validated the splicing defects in three previously characterized genes. We showed that transcripts of the flowering repressor gene FLC and the cell division inhibition gene ICK2 [53] have intron splicing defects (Fig 3E-3I), and transcripts of the cell cycle promotion gene CYCD2;1 [54] have a skipped exon in emb1579 (Fig 3J-3L). The defects in global gene transcription and pre-mRNA splicing provide an explanation for why emb1579 mutants exhibit pleiotropic growth and developmental defects. In support of the notion that the gene transcription and pre-mRNA splicing defects account for the developmental defects in emb1579 mutants, we found that down-regulation of FLC, which has both transcriptional and mRNA splicing defects in emb1579 mutants (Fig 3C, 3E and 3F), can suppress the later flowering phenotype in emb1579 mutants (S5 Fig). In summary, we demonstrate that EMB1579 is involved in the regulation of transcription and pre-mRNA splicing in Arabidopsis.

Some proteins crucial for gene transcription and mRNA splicing colocalize with EMB1579 condensates in the nucleus
We next determined whether specific proteins crucial for gene transcription and mRNA splicing can enter EMB1579 condensates in vivo. We found that MSI4 and DDB1A appeared in the EMB1579 condensates pull-down fraction (Fig 3A, S1 Table). MSI4/FVE is known to be a key regulator of the autonomous flowering pathway that constitutively reduces the expression of FLC [36]. MSI4 interacts with CUL4-DDB1 and a PRC2-like complex to mediate this function [36]. Therefore, we performed colocalization experiments to determine whether MSI4 and DDB1B (homologue of DDB1A) can enter EMB1579 condensates. Both proteins do indeed colocalize with EMB1579 condensates in vivo ( Fig 4A). Spliceosome components, including uridine-rich small nuclear ribonucleoproteins (U snRNPs), non-snRNP factors, serine/ recombinant EMB1579 protein. The asterisks indicate EMB1579 protein bands. The lower band might be a degradation product of the full-length EMB1579. The original pictures are available in S1 Raw Images.   Table. (B) EMB1579-activated and EMB1579-repressed genes. Seven-DAG WT and emb1579 seedlings were subjected to RNAseq analysis. A full list of the up-and down-regulated genes is presented in S2 Table. (C) Validation of the altered expression of specific genes in emb1579 mutants. qRT-PCR was performed to confirm the changed expression levels of 17 different genes in emb1579 mutants, as originally revealed by RNA-seq analysis. Data are presented as mean ± s.e.m, arginine (RS) proteins, and heterogeneous nuclear ribonucleoproteins (hnRNPs), are known to play important roles in mRNA splicing [5,55,56]. We selected U1-70K and SC35 as representative U1 snRNP and non-snRNP factors [57][58][59], respectively, and found that they colocalize with EMB1579 condensates (Fig 4A). RZ-1C, the hnRNP-like RNA binding protein (RBP) in plants, was reported to be involved in the regulation of mRNA splicing [5,11]. We examined the spatial association of RZ-1C with EMB1579 and found that they colocalize in cells ( Fig  4A). Thus, we demonstrate that interactors of PRC2 and some spliceosome factors colocalize with EMB1579 condensates in vivo. To determine whether the colocalized or pulled-down proteins physically interact with EMB1579, we performed the firefly split luciferase complementation imaging assay and found that 18 of them do indeed interact with EMB1579 ( Fig  4B). Taking these data together, we propose that EMB1579 undergoes LLPS to form liquid-like condensates that condense biomolecules crucial for chromosomal function and RNA biology to control nuclear events, such as transcription and pre-mRNA splicing ( Fig 4C).

The RED repeat is required for the formation of normal-sized EMB1579 condensates and the cellular functions of EMB1579
To directly link the formation of EMB1579 condensates with the functions of EMB1579 in vivo, we wanted to identify the motifs within EMB1579 that facilitate its LLPS and formation of liquid-like condensates. The RED repeat in EMB1579 (Fig 5A) caught our attention because some proteins containing RE, RD, or RED repeats were previously shown to form bodies in cells [60][61][62]. To determine whether the RED repeat contributes to the formation of EMB1579 condensates, we deleted the RED repeat from the EMB1579 protein to create EMB1579ΔRED (Fig 5B). At all the concentrations tested, EMB1579ΔRED condensates are significantly smaller than EMB1579 condensates in vitro (Fig 5C and 5D), which suggests that the RED repeat is required for EMB1579 to form normal-sized condensates. Furthermore, a higher concentration of EMB1579ΔRED is required to form liquid condensates compared to EMB1579 at the same salt concentration (Fig 5E), which suggests that the RED repeat might modulate the interaction between EMB1579 molecules. Interestingly, we found that EMB1579ΔRED condensates are even more dynamic than EMB1579 condensates (Fig 5F and 5G). Although we do not currently know the reason for this, it could be that the RED repeat, which is enriched in n = 3. The selected genes were demonstrated previously to be involved in different physiological processes, as indicated in (B). The underlying numerical data are available in S1 Data. (D) Summary of splicing defects events in emb1579 mutant plants. A full list of the splicing defects is presented in S3 Table. Fig 3A) and three colocalized proteins directly interact with EMB1579. (C) Simple model for the phase separation of EMB1579 in the nucleus and its potential functions in Arabidopsis. EMB1579 undergoes phase separation to form dynamic compartments in the nucleus. The charged residues, strengthens the electrostatic interaction between EMB1579 molecules ( Fig  5A). Nonetheless, these data together suggest that the RED repeat modulates the formation of normal-sized EMB1579 condensates in vitro. We next introduced EMB1579ΔRED-TGFP into emb1579 mutants with its expression under the control of the native EMB1579 promoter and selected the transgenic plants contain similar amount of EMB1579ΔRED-TGFP when compared to EMB1579-TGFP for subsequent analyses (S6A and S6B Fig). We found that EMB1579-TGFP clearly formed condensates whereas EMB1579ΔRED-TGFP hardly formed any large condensates in cells visualized by laser scanning confocal microscopy ( Fig 5H and  5I) or superresolution structured illumination microscopy (SIM) (Fig 5J). This suggests that the RED motif is required for EMB1579 to form normal-sized condensates in vivo. Consistent with this, we found that EMB1579ΔRED-TGFP only partially restored the transcript levels of several genes compared to EMB1579-TGFP ( Fig 5K). We also found that EMB1579ΔRED-TGFP, in contrast to EMB1579-TGFP, failed to rescue the primary root growth and floral transition phenotypes in emb1579 mutants (S6C- S6F Fig). However, we found that EMB1579ΔRED retained the capability to interact with the functionally relevant interactors (S7 Fig). These data together suggest that the cellular functions of EMB1579 likely depend on the formation of appropriate liquid-like compartments.

EMB1579 interacts with MSI4 and emb1579 phenocopies msi4 in terms of flowering time and the level of H3K27m3 on FLC
Above, we showed that MSI4 colocalizes with EMB1579 condensates in the nucleus ( Fig 4A) and EMB1579 physically interacts with MSI4 ( Fig 4B). To understand how exactly EMB1579 performs its cellular functions, we wanted to determine how EMB1579 coordinates with MSI4 to control the expression of FLC. The physical interaction between EMB1579 and MSI4 was further confirmed by yeast two-hybrid assay ( Fig 6A). Next, we found that although both emb1579 and msi4 mutants exhibit a late-flowering phenotype ( Fig 1H and 1I), this phenotype is weaker in emb1579 than in msi4 (Fig 6B and 6C). This allows us to speculate that EMB1579 may act as a regulator of MSI4 in controlling the expression of FLC and flowering. It was previously proposed that MSI4 acts by interacting with a PRC2-like complex to establish and/or maintain the levels of H3K27me3 at the FLC locus [36]. Considering that the global level of H3K27me3 is reduced in msi4 mutants [36], we determined the global level of H3K27me3 in emb1579 mutants and found that it is reduced ( Fig 6D). To specifically link the function of EMB1579 to the expression of FLC, we analyzed the level of H3K27me3 on FLC and found that it is reduced in emb1579 mutants ( Fig 6E). Again, this is in agreement with the previous report that msi4 mutants have reduced levels of H3K27me3 on FLC [36]. Thus, we showed that EMB1579 physically interacts with MSI4 and emb1579 phenocopies msi4 in terms of flowering time and the level of H3K27me3 on the FLC locus.

EMB1579 condensates recruit and condense MSI4 in vitro and in vivo
Above, we showed that deletion of the RED repeat of EMB1579 impairs the repression of FLC transcription ( Fig 5K) and affects flowering time (S6E Fig). These data suggest that the formation of normal-sized liquid-like condensates is crucial for this functional role of EMB1579. Considering that MSI4 colocalizes with EMB1579 condensates in the nucleus (Fig 4A)

PLOS BIOLOGY
Phase separation of EMB1579 controls transcription and mRNA splicing in plants . Data are presented as mean ± s.e.m. ��� P < 0.001 by Student t test. Numerical data underlying this panel are available in S1 Data. (E) The capability of EMB1579ΔRED to form condensates is impaired in vitro. The upper panel shows the phase diagram of EMB1579ΔRED condensate formation at the indicated protein and KCl concentrations. The phase diagram was plotted as described in Fig 2L. Red circles indicate the formation of droplets; black squares indicate no formation of condensates. The blue line indicates the phase boundary of EMB1579ΔRED. The lower panel shows the saturation curves for the phase separation of EMB1579 and EMB1579ΔRED. The protein concentration in the dilute phase (solid circles) is plotted for various total protein concentrations (empty circles) at five different KCl into the condensates. Indeed, we found that EMB1579 condensates can recruit and condense recombinant MSI4 protein in vitro (Fig 6F and 6G). Next, we incubated EMB1579 with Arabidopsis root total extract containing red fluorescent protein (RFP)-MSI4 under the conditions that induce phase separation of EMB1579 in vitro. We found that EMB1579 condensates actively recruit and condense RFP-MSI4 in this system (Fig 6H), which suggests that EMB1579 condensates might be able to condense MSI4 in vivo. In support of this speculation, we found that loss of function of EMB1579 impaired the capability of RFP-MSI4 to form compartments in Arabidopsis protoplasts ( Fig 6I). These data together suggest that EMB1579 condensates can recruit and condense MSI4.

EMB1579 interacts with CUL4 and DDB1 but not CLF and FIE
MSI4 regulates the flowering time by interacting with CUL4-DDB1 [36]. Furthermore, MSI4 and DDB1 appeared in the EMB1579 pull-down fractions (Fig 3A, S1 Table), and they both colocalized with EMB1579 condensates in vivo (Fig 4A). This prompted us to ask how EMB1579 may functionally coordinate with CUL4-DDB1 MSI4 complex to maintain the level of H3K27me3 on FLC and repress the expression of FLC. We used the firefly split luciferase complementation imaging assay to show that EMB1579 interacts with CUL4 in planta (S8A Fig). This result, together with our finding that EMB1579 interacts with MSI4 and DDB1B in planta (Fig 4B), suggests that EMB1579 interacts with MSI4, DDB1, and CUL4. The interactions were further confirmed by the yeast two-hybrid assay. Specifically, we found that MSI4 can interact with full-length EMB1579 (S8B Fig), and a fragment of EMB1579 containing the MSI4-binding domain can bind to CUL4 (S8C and S8D Fig). Interestingly, we found that the RED repeat is not required for the interaction between EMB1579 and MSI4 (S8C Fig). MSI4 regulates the flowering time by forming CUL4-DDB1 MSI4 complex that interacts with a CLF-PRC2 complex to repress the expression of FLC via H3K27 methylation [36], and loss of function of EMB1579 reduces the level of H3K27me3 on FLC chromatin. Therefore, we wondered whether EMB1579 physically interacts with CLF and FIE, core subunits of CLF-PRC2. We found that EMB1579 did not interact with CLF and FIE (S8A, S8B and S8D Fig). These data together suggest that EMB1579 regulates the level of H3K27me3 on FLC chromatin via interaction with CUL4-DDB1 MSI4 complex rather than direct interaction with CLF-PRC2. As EMB1579 condensates can recruit and condense MSI4 both in vitro and in vivo, we wondered whether EMB1579 condensates can also recruit and condense DDB1 and CUL4. We thus performed the in vitro recruitment experiments using DDB1B as the representative protein, since it colocalizes with EMB1579 condensates in vivo (Fig 4A). We found that EMB1579 condensates can concentrations for EMB1579 and EMB1579ΔRED. The same colored solid and empty circles represent the data from the same set of experiments. The concentration of the dilute phase for EMB1579 and EMB1579ΔRED falls directly onto the phase boundary of EMB1579 and EMB1579ΔRED (solid lines) for all conditions. The underlying numerical data are available in S1 Data. (F) FRAP analysis of EMB1579 and EMB1579ΔRED condensates. Bars = 2 μm. (G) Plot of the fluorescence intensity during FRAP experiments. The blue curve shows the average fluorescence intensity from 15 EMB1579 bodies and the red curve represents the average fluorescence intensity from 20 EMB1579ΔRED bodies. Data are presented as the mean. The dashed lines represent t 1/2 of EMB1579 and EMB1579ΔRED. Numerical data underlying this panel are available in S1 Data. (H) No obvious compartments are formed by EMB1579ΔRED-TGFP compared to EMB1579-TGFP. GFP-NLS is a control that shows diffuse fluorescence in the nucleoplasm. Bar = 20 μm for the whole image and bar = 5 μm for the inset image. (I) Analysis of the coefficient of variation of fluorescence of GFP, EMB1579-TGFP, and EMB1579ΔRED-TGFP. Data are presented as mean ± s.e.m. ��� P < 0.001 by Student t test. More than 30 nuclei were measured from 20 seedlings. Numerical data underlying this panel are available in S1 Data. (J) SIM images of nuclei harboring EMB1579-TGFP or EMB1579ΔRED-TGFP in Arabidopsis root cells. The bodies formed by EMB1579-TGFP and EMB1579ΔRED-TGFP are indicated by blue and red arrows, respectively. Bar = 5 μm. The gray values of EMB1579-TGFP (blue) and EMB1579ΔRED-TGFP (red) were measured along the lines shown in the left panel. The triangles indicate the peaks of fluorescence. The underlying numerical data are available in S1 Data. (K) qRT-PCR analysis to determine the transcript levels of six genes in WT, emb1579, and complementation lines. Data are presented as mean ± s.e.m, n = 3. The underlying numerical data are available in S1 Data. aa, amino acid; EMB1579, EMBRYO DEFECTIVE 1579; FRAP, fluorescence recovery after photobleaching; GFP, green fluorescent protein; LLPS, liquid-liquid phase separation; NLS, nuclear localization signal; qRT-PCR, quantitative reverse transcription PCR; SIM, structured illumination microscopy; TGFP, tandem copies of enhanced GFP; WT, wild type.
https://doi.org/10.1371/journal.pbio.3000782.g005 indeed recruit and condense DDB1B in vitro (S9 Fig). Therefore, we speculate that EMB1579 performs its cellular functions by forming condensates that can recruit and condense CUL4-DDB1 MSI4 complex to increase their local concentration. This will facilitate the interaction of CUL4-DDB1 MSI4 complex with CLF-PRC2 complex to promote its role in establishing and/or maintaining the level of H3K27me3 on FLC, thereby repressing FLC transcription and promoting flowering (Fig 6J).

Discussion
We here found that EMB1579 is involved in numerous physiological processes and is crucial for Arabidopsis growth and development. We demonstrate that EMB1579 forms liquid-like condensates in vivo and is able to undergo LLPS in vitro. We provided direct evidence that the RED repeat is important for EMB1579 to form normal liquid-like condensates and is crucial for EMB1579 to perform its physiological functions. Using floral transition as the representative physiological process, we demonstrate that EMB1579 is involved in FLC-mediated repression of flowering, presumably by interacting with MSI4, CUL4, and DDB1 and condensing them through the formation of liquid-like condensates. Consequently, EMB1579 might facilitate the interaction of CUL4-DDB1 MSI4 complex with CLF-PRC2 to promote its role in establishing and/or maintaining the level of H3K27me3 on FLC, thereby repressing the expression of FLC and promoting flowering in Arabidopsis. This work provides insights into both plant development regulatory programs and liquid-like biological systems and raises many intriguing questions for future research.

EMB1579 compartments are highly dynamic phase-separated condensates
Consistent with the presence of IDRs in EMB1579 (Fig 2I), we found that EMB1579 undergoes LLPS in vitro (Fig 2K). EMB1579 compartments are roughly spherical and fuse with each other both in vitro and in vivo, and they undergo rapid internal rearrangement (Fig 2F-2H, Fig 2M-2O). These characteristics support the notion that EMB1579 compartments are liquid-like condensates. Strikingly, our data suggest that EMB1579 compartments are extremely dynamic, as the half-time of recovery of EMB1579 compartments was determined to be 7.4 seconds in vivo ( Fig 2G) and 2.61 seconds in vitro (Fig 2N). The extraordinarily dynamic properties of EMB1579 compartments might be due to the fact that the EMB1579 protein is mostly occupied by IDRs (Fig 2I), which do not adopt stable secondary and tertiary structures and are conformationally heterogeneous and dynamic. Although EMB1579 is conserved in higher plants (S10 Fig), no full-length homologues of EMB1579 have been found in mammalian systems. Therefore, no side-by-side comparison can be performed in terms of the LLPS dynamics of EMB1579 condensates. However, EMB1579 condensates are much more dynamic than some previously reported subnuclear condensates in mammalian systems. For instance, the half-time of recovery is about 23 seconds for nucleophosmin (NPM1) [63] and 19-64 seconds for the RBP hnRNPA1 [64]. It was reported that cytoplasmic elements in stationary plant cells (e.g., cortical microtubules and the actin cytoskeleton [65,66]) are more dynamic than their counterparts in mammalian systems, and this was proposed to be an adaptation strategy for sessile plants. In this regard, it might be fair for us to speculate that the extraordinarily dynamic EMB1579 condensates enable sessile plants to rapidly alter their DNA-and RNA-based activities to control transcription and pre-mRNA splicing in response to external and internal stimuli.

The RED repeat modulates the formation of EMB1579 condensates, and it is required for EMB1579 to perform cellular functions
More and more proteins have been shown to undergo LLPS and participate in numerous fundamental physiological processes [19,67]. For most of these proteins, however, the direct link between the formation of condensates and their physiological functions still remains to be established. Here, we specifically identify that the RED repeat is important for the formation of normal-sized EMB1579 condensates in vitro and in vivo (Fig 5C-5J), and we demonstrate that the RED repeat is required for EMB1579 to perform cellular functions (Fig 5K, S6C-S6F Fig). Our finding that alternating basic-acidic dipeptide repeats are required for the formation of functionally normal EMB1579 condensates enhances our understanding of the principles that govern the in vivo dynamics of liquid-like bodies. Some other proteins containing RED, ER, or RD repeats also form subnuclear compartments in cells [60][61][62] and have been implicated in fundamental physiological processes. It is interesting to speculate whether the basic-acidic dipeptide repeats have a conserved function in LLPS of those proteins and formation of functionally competent condensates. Our study lays a foundation for further deep dissection of the mechanism of action of proteins containing RED, ER, or RD alternating basic-acidic dipeptide repeats.

EMB1579 compartments condense different biomolecules and, specifically, they regulate the function of MSI4
The nature of the functions that are performed by nonmembranous compartments and how those functions are performed are two fundamental questions in the phase-separation field. Consistent with our results that many nuclear proteins crucial for DNA and RNA biology are pulled down by EMB1579 (Fig 3A, S1 Table), we found that global gene transcription and mRNA splicing were altered in emb1579 mutants (Fig 3B and 3C, S2 and S3 Tables). This explains why loss of function of EMB1579 causes pleiotropic developmental defects in Arabidopsis (Fig 1). Importantly, some proteins crucial for gene transcription and pre-mRNA splicing colocalize with EMB1579 condensates in vivo (Fig 4A), which allows us to speculate that EMB1579 condensates regulate gene transcription and mRNA splicing by recruiting and condensing those proteins (Fig 4C). We showed that MSI4 appears in the EMB1579 pull-down fraction (Fig 3A, S1 Table), and emb1579 mutants have similar FLC-related phenotypes to msi4 (Figs 3C, 6B, 6C and 6E). Therefore, to understand how EMB1579 performs its physiological functions, we further investigated the functional relationship between EMB1579 and MSI4 in the FLC-mediated repression of flowering. MSI4 interacts with CUL4-DDB1 and CLF-PRC2 [36], and we demonstrated that EMB1579 is involved in the transcription of FLC by interacting with CUL4-DDB1 MSI4 complex but not CLF-PRC2. Based on our results, we propose that EMB1579 functions by forming condensates that recruit and condense CUL4-DDB1 MSI4 complex to facilitate the interaction of this complex with CLF-PRC2, which in turn controls the level of H3K27me3 on FLC and the transcription of FLC to regulate flowering (Fig 6J). Condensation of the interactors of PRC2 through subcompartmentalization may provide a way to regulate the function of PRC2 in establishing and/or maintaining the level of H3K27me3 on chromatin. In particular, condensation of the scaffold protein MSI4 may facilitate the assembly of functional PRC2 complexes, although it was shown that MSI4 is not the core subunit of EMF complex [35]. Considering that CUL4 interacts with FLC chromatin in a MSI4-dependent manner [36], the condensation of CUL4-DDB1 MSI4 complex might facilitate the targeting and/or recruitment of PRC2 to chromatin. Our study enriches our understanding of transcriptional regulation via PRC2, a major regulator of specific gene expression patterns and developmental programs in different organisms. Interestingly, it was shown that some PcG complex proteins can undergo LLPS and form liquid-like condensates in the nucleus [68,69]. It will be interesting to investigate how EMB1579 condensates function coordinately with liquid-like condensates formed by PcG complex proteins, if plant PcG complex proteins can also undergo LLPS. Nonetheless, our study significantly enhances our understanding of how LLPS systems perform cellular functions.
These data together allow us to propose that EMB1579 controls Arabidopsis growth and development by forming nonmembranous compartments that recruit and condense important biomolecules to drive complex biochemical reactions. Two important questions for future research are how EMB1579 condensates selectively condense different players and how EMB1579 condensates sense intrinsic and environmental cues to adjust their dynamic properties to meet different physiological demands. It was proposed that EMB1579 is involved in nuclear calcium signaling during the salt response, since it contains an EF-hand motif [44]. It will be interesting to investigate whether EMB1579 condensates can sense nuclear calcium signaling through the EF-hand to adjust their phase-separation dynamics. In addition, MAP190, the tobacco homologue of EMB1579, can interact with actin filaments and microtubules [43], and cytoskeletal systems are able to phase separate [70,71]. In the future, it will be important to determine how LLPS of EMB1579 links the nuclear skeleton to gene transcription and mRNA splicing.

Plant materials and growth conditions
The T-DNA insertion lines CS16026 and Salk_007142 were designated as emb1579-1 and emb1579-3, respectively. The mutant mis4 (Sail_1167_E05) has been described previously [36]. The genotyping of T-DNA insertion mutants was performed with the primers listed in S4 Table. Arabidopsis Columbia-0 (Col-0) ecotype was used as WT, and plants were grown in soil or media under a 16-hour-light/8-hour-dark photoperiod at 22˚C.

Construction of plasmids
The promoter of EMB1579, a 2,278-bp DNA fragment upstream of the initiation ATG of EMB1579, was amplified with primers proEMB1579-SalI-F and proEMB1579-BamHI-R (S4 Table) from Arabidopsis genomic DNA and subsequently moved into pCAMBIA1301 to generate pCAMBIA1301-proEMB1579. The genomic sequence of EMB1579 (gEMB1579, 6,686 bp) was amplified with primers gEMB1579-SmaI-F and gEMB1579-SacI-R (S4 Table) using Arabidopsis genomic DNA as the template and cloned into pEASY-Blunt to generate pEASY-Blunt-gEMB1579. Three TGFP and gEMB1579 were sequentially moved into pCAMBIA1301-proEMB1579 to generate pCAMBIA1301-proEMB1579::gEMB1579-TGFP. To delete the RED repeat flanking the sequence from 1461 to 1694 in gEMB1579, a PCR fragment was amplified with primers ΔRED-F and ΔRED-R (S4 Table) using the pEASY-Blunt-gEMB1579 plasmid as the template and moved into pCAMBIA1301-proEMB1579-TGFP digested with SmaI and SacI to generate pCAMBIA1301-proEMB1579::gEMB1579ΔRED-TGFP.
To investigate the tissue expression pattern of EMB1579, the region containing the promoter and the first two exons of EMB1579 was amplified with primers proEMB1579-SalI-F and GUS-BamHI-R (S4 Table) using Arabidopsis genomic DNA as the template. The amplified PCR product was cloned into pBI101, which contains the GUS coding region, to generate pBI101-EMB1579pro::GUS. To observe the fluorescence heterogeneity of EMB1579ΔRED-TGFP, pCAMBIA1301-35S::GFP-NLS-NOS was constructed as a control. The promoter of 35S was amplified with primer pair 35S-HindIII-F/35S-PstI-R (S4 Table), using pFGC5941 as the template, and subsequently cloned into pCAMBIA1301 to generate pCAMBIA1301-35S. GFP-NLS and NOS were amplified with primer pairs GFP-NLS-SacI-F/GFP-NLS-EcoRI-R and NOS-EcoRI-F/NOS-BstXI-R (S4 Table) using pEZS-NL and pBINPLUS-35S::U2B@-RFP-NOS as the template, respectively. These two fragments were cloned into pCAMBIA1301-35S to generate pCAMBIA1301-35S::GFP-NLS-NOS.
To determine whether proteins of interest can physically interact with EMB1579 or EMB1579ΔRED, the coding regions of 23 genes were amplified with the following primer pairs: To determine the intracellular localization of MSI4 in WT and emb1579 Arabidopsis protoplasts, the CDS of MSI4 was amplified from cDNA of Arabidopsis seedlings with primer pair MSI4-KpnI-F2/MSI4-XbaI-R (S4 Table), and mRFP was amplified with primer pair RFP-Sa-lI-F/RFP-KpnI-R (S4 Table) for the fusion with MSI4. MSI4 and mRFP were subsequently moved into pEZS-NL to generate pEZS-NL-mRFP-MSI4.
To generate an FLC-RNAi construct, the sequence containing the second and third exons of FLC was amplified with primers FLC-RNAi-XbaI-AscI-F containing XbaI and AscI sites and FLC-RNAi-BamHI-SwaI-R containing BamHI and SwaI sites (S4 Table), using Arabidopsis seedlings' cDNA as the template. The amplified inverted-repeat PCR fragment was subsequently digested with AscI/SwaI or XbaI/BamHI to generate two complementary RNAi fragments, which were moved into pFGC5941 to generate the pFGC5941-FLC-RNAi construct.
To generate the recombinant GFP protein as the control for determining the concentration of EMB1579 in the nucleus, GFP was amplified with primer pair GFP-SalI-F/GFP-XbaI-R (S4 Table) and moved into pColdI digested with SalI and XbaI to generate pColdI-GFP.

Transient expression in Arabidopsis mesophyll protoplasts
Arabidopsis mesophyll protoplasts were isolated from WT or emb1579 Arabidopsis plants according to the published method [73]. The plasmid pEZS-NL-RFP-MSI4 was introduced into protoplasts derived from WT or emb1579 plants via the PEG-calcium-mediated transformation method [73]. The transformed protoplasts were observed under an Olympus FV1200MPE laser scanning confocal microscope 8-12 hours after transformation. mRFP was excited with Argon 561-nm laser line.

Firefly split luciferase complementation imaging assay
The A. tumefaciens strain GV3101 was cultured in LB liquid medium with 50 μg/ml kanamycin and 50 μg/ml rifampicin at 28˚C overnight. Bacterial cells were pelleted under 4,000g for 10 minutes at room temperature and subsequently resuspended with 10 mM MES (pH 5.6), 10 mM MgCl 2 , 100 μM acetosyringone to a final concentration of OD 600 = 0.8. After 2-4 hours' incubation at room temperature, the strains containing different plasmids were mixed equivoluminally and infiltrated into 3-week-old N. benthamiana leaves. After 60 hours of infiltration, the leaves were sprayed with 1 mM D-luciferin and the signal was detected with a NightShade LB 985 plant imaging system.

Transient expression in N. benthamiana
A. tumefaciens strain GV3101 was transformed with pCAMBIA1301-35S::EMB1579-GFP-NOS, pFGC5941-RFP-MSI4, and p19. The three transformants were mixed equivoluminally and infiltrated into 3-week-old N. benthamiana leaves. The procedures and buffers were prepared as previously described [74]. After 60 hours of infiltration, the leaf epidermal cells were observed under an Olympus FV1200MPE laser scanning confocal microscope using 488-nm excitation for GFP and 561-nm excitation for RFP.

Particle bombardment
Transient expression of GFP-NLS in the leaves of 3-to 4-week-old WT Arabidopsis plants was achieved using a biolistic DNA delivery system. The procedure of bombardment of leaves using 35S::GFP-NLS-NOS plasmid was performed according to the published method [74]. After bombardment, the leaves were placed on MS medium with 0.8% agar and incubated at 23˚C for 48 hours in the dark. Leaves were observed at 488-nm excitation for GFP fluorescence.

Yeast two-hybrid assay
The bait and prey plasmids were cotransformed into Saccharomyces cerevisiae reporter strain Gold competent cells according to the manufacturer's instructions (Clontech). After growing on SD/-Leu/-Trp medium at 30˚C for 2-3 days, the transformants were selected on SD/-His/-Leu/-Trp/-Ade medium at 30˚C for 3 days to detection interactions. White yeast that grows well on SD/-His/-Leu/-Trp/-Ade medium indicates a positive interaction.

RT-PCR analysis
Total RNA was extracted from 2-week-old Arabidopsis seedlings. After treatment with DNase I (M0303S, NEB), first-strand cDNA was synthesized with 10 μg total RNA using M-MLV Reverse Transcriptase (M170A, Promega) and oligo d(T) 18 (3806, TaKaRa) primer or random primer (for mRNA splicing detection). The truncated EMB1579 transcripts in emb1579-1 and emb1579-3 were detected by semiquantitative RT-PCR using three primer pairs as follows: for emb1579-1, EMB1579-C F -RT/EMB1579-C R -RT, EMB1579-D F -RT/EMB1579-D R -RT, and EMB1579-E F -RT/EMB1579-D R -RT; for emb1579-3, EMB1579-A F -RT/EMB1579-A R -RT, EMB1579-B F -RT/EMB1579-B R -RT, and EMB1579-C F -RT/EMB1579-C R -RT (S4 Table). The full-length transcripts of EMB1579 were amplified with primer pair EMB1579-qRT-F2/ EMB1579-qRT-R2 (S4 Table). eIF4A was amplified with primer pair eIF4A-F1/eIF4A-R1 (S4 Table) as the internal control. The transcript level of EMB1579 was also detected with qRT-PCR using primer pair EMB1579-qRT-F1/EMB1579-qRT-R1 (S4 Table). eIF4A was amplified with primer pair eIF4A-F2/eIF4A-R2 (S4 Table) as the internal control. The relative mRNA expression levels in samples were calculated according to the 2 -44Ct method [75]. qRT-PCR was performed with primers FLC intron1 SF and FLC intron1 SR (S4 Table) to detect the expression level of FLC in WT and emb1579 plants. To detect the splicing status (spliced or unspliced) of the first intron of FLC, PCR products were amplified with primer pairs FLC intron1 UF/FLC intron1 UR and FLC intron1 SF/FLC intron1 SR (S4 Table) using cDNA from WT and emb1579 seedlings by random primed reverse transcription [11]. To determine the splicing status (spliced or unspliced) of the third intron of ICK2, primer pairs ICK2-SF/ICK2-SR and ICK2-UF/ICK2-UR (S4 Table) were used, respectively. For CYCD2;1, primer pairs CYCD2;1-SF/CYCD2;1-SR and CYCD2;1-UF/CYCD2;1-UR (S4 Table) were used to measure the levels of abnormal CYCD2;1 transcripts, which lack the second and third exons, and normal CYCD2;1 transcripts, which contain the second and third exons. The splicing efficiency was determined by normalizing the level of spliced transcripts to the level of unspliced transcripts for each splicing defect as described previously [11,76]. qRT-PCR was conducted using 2×RealStar Green Power Mixture (with ROX II) in an Applied Biosystems 7500 Fast Real-Time PCR System. All primers used in the qRT-PCR for the validation of RNA-seq data are listed in S5 Table. All data were obtained from three biological replicates.

Determination of the concentration of EMB1579 in the nucleus
Briefly, 2-week-old proEMB1579::gEMB1579-TGFP; emb1579 Arabidopsis seedlings (approximately 3 g) were ground in liquid nitrogen. Subsequently, a certain amount (V1 mL) of extraction buffer (20 mM Tris-HCl [pH 7.4], 25% glycerol, 20 mM KCl, 2 mM EDTA, 2.5 mM MgCl 2 , 250 mM sucrose, 5 mM DTT), supplemented with complete EDTA-free Protease Inhibitor Cocktail, was added into the ground mixture. After centrifugation at 10,000g for 10 minutes at 4˚C, the supernatant at V2 mL was collected. Next, the nuclei were isolated and total nuclear proteins were extracted as described previously [77]. The amount of EMB1579-TGFP (ng) in the total nuclear protein was determined by western blot analysis probed with anti-GFP antibody, using a known amount of recombinant GFP protein (1-10 ng) as the control. To calculate the molar concentration of EMB1579-TGFP in the nucleus, we initially assumed that the volume of cells is equal to (V1 − V2) mL and the nucleus occupies roughly 2.6% of the whole cell volume, as reported previously [80]. The amount of EMB1579-TGFP (ng) was finally divided by the volume of nuclei ([V1 − V2] � 2.6%) and the molecular weight of EMB1579-TGFP (260 kDa) to yield the molar concentration of EMB1579-TGFP (nM) in the nucleus, as described previously [81].

GUS staining
The seedlings were immersed in the GUS staining solution (0.5 mg/ml X-glucuronide in 100 mM sodium phosphate [pH 7.0], 0.5 mM ferricyanide, 0.05 mM ferrocyanide, 1 mM EDTA, and 0.1% Triton X-100) for 14 hours at 37˚C. GUS-stained materials were cleared of chlorophyll in 70% (v/v) ethanol. To perform GUS staining in seeds, the siliques were dissected to expose seeds. The seeds were initially incubated with GUS staining solution for 20 minutes under vacuum infiltration, followed by an additional incubation for 4 hours at 37˚C. Finally, the embryos were cleared by HCG solution (eight parts chloroacetaldehyde: three parts water: one part glycerol). Stained materials were photographed under an Olympus IX83 microscope.

Imaging of Arabidopsis root and embryo cells
To image Arabidopsis root cells and embryonic cells, propidium iodide (PI) staining was performed. Briefly, Arabidopsis roots were incubated with PI at 20 μg/ml for about 5 minutes. After rinsing with ddH 2 O, they were mounted in ddH 2 O for visualization. The staining of embryos with PI was performed according to a previously published method [45]. Briefly, embryos were isolated from ovules and subsequently incubated with PI solution (50 μg/ml; dissolved in 9% glucose and 5% glycerol) for about 10 minutes before observation. Arabidopsis roots and embryos were observed under an Olympus FV1200MPE laser scanning confocal microscope with the excitation wavelength set at 543 nm for PI staining. To visualize Arabidopsis embryos at earlier stages, the ovules were dissected and cleared in Hoyer's solution [82] from 2 hours to overnight depending on developmental stage and were subsequently observed under an Olympus BX53 microscope equipped with differential interference contrast (DIC) optics.

Complementation of emb1579 and visualization of the intracellular localization and dynamics of EMB1579
To visualize the intracellular localization of EMB1579-TGFP or EMB1579ΔRED-TGFP, roots from pCAMBIA1301-proEMB1579::gEMB1579-TGFP; emb1579 seedlings or pCAM-BIA1301-proEMB1579::gEMB1579ΔRED-TGFP; emb1579 seedlings that were vertically cultured on petri dishes for 2-3 days were observed under a fluorescence laser scanning confocal microscope equipped with a ×100 oil objective (numerical aperture of 1.4) at 488-nm excitation. The Z-series images were collected with the step size set at 0.5 μm. The nuclei were stained with Hoechst 33342 (30 μg/ml), which was excited with a 405-nm laser.
To capture the dynamics of EMB1579-TGFP during the cell cycle, the sample was visualized by a spinning disk confocal microscope, and the time-series Z-stack images were collected at 2-minute intervals with the step size set at 1 μm. The collected images were processed with ImageJ software. To reveal the internal dynamics of EMB1579 bodies, FRAP was carried out with an Olympus FV1200MPE laser scanning confocal microscope. Two scan images were captured before bleaching and a single body was bleached using 100% power with 488-nm and 10% power with 405-nm laser lines for 5 seconds. Time-lapse images were collected at 4-second intervals. The gray values of single bodies were analyzed with ImageJ software against the elapsed time to calculate the rate of fluorescence recovery. The fluorescence intensity of prebleaching was normalized to 100%. Tracking of body fusion events was performed in root meristem cells. The difference between EMB1579-TGFP and EMB1579ΔRED-TGFP in forming subnuclear compartments was assessed by measuring the coefficient of variation of GFP fluorescence in the nucleoplasm with ImageJ software as described previously [83].

Protein production
To generate recombinant EMB1579 and EMB1579ΔRED proteins, the constructs pFastBa-c1-EMB1579 and pFastBac1-EMB1579ΔRED were transformed into DH10Bac competent cells and screened by blue/white selection with 100 μg/ml Bluo-gal, 40 μg/ml IPTG, 50 μg/ml kanamycin, 7 μg/ml gentamycin, and 10 μg/ml tetracycline in LB agar plates. The recombinant bacmids extracted from the white colonies were transfected into Sf9 insect cells with Cellfectin II Reagent. After two generations of virus propagation, the virus was inoculated into Sf9 insect cells for large-scale protein production. The transfected cells were collected at 1,000g for 10 minutes under 4˚C, suspended in binding buffer (

Preparation of fluorescently labeled EMB1579 and its variants
To allow direct visualization of phase separation by fluorescence light microscopy, purified recombinant EMB1579 and EMB1579ΔRED proteins dialyzed against buffer B (25 mM Hepes [pH 8.0], 250 mM KCl) were subjected to fluorescent labeling with Oregon Green (see below). After the proteins were allowed to undergo phase separation in F-buffer (25 mM Hepes [pH 8.0], 100 mM KCl, 100 mg/ml PEG 3350) for 1 hour at room temperature, an 8-fold molar ratio of Oregon Green was added and the mixture was further incubated in F-buffer for 90 minutes at room temperature. The reaction mixtures were centrifuged under 15,000g for 30 minutes at room temperature, and the pellet was washed with F-buffer twice to remove the free Oregon Green and subsequently resuspended in buffer E (25 mM Hepes [pH 8.0], 500 mM KCl). The suspension was clarified under 15,000g at 4˚C for 30 minutes, and the supernatant was used for the subsequent phase-separation experiments.

Phase transition of EMB1579 and its variants in vitro
EMB1579 and its variants were preclarified under 14,000g for 20 minutes at 4˚C before the reaction. To allow the phase separation to occur in vitro, a range of EMB1579 concentrations from 0.01 μM to 1 μM and a range of EMB1579ΔRED concentrations from 0.05 μM to 1 μM were incubated in P buffer (25 mM Tris-HCl [pH 8.0], 100-2,000 mM KCl, 2 mM DTT). The phase separation of EMB1579 and its variants was directly visualized by either DIC optics or fluorescent light microscopy. To monitor the dynamic phase-separation process, the protein mixture in P buffer with 100 mg/ml PEG 3350 was injected into a flow chamber with the coverslip pretreated with Poly-L-lysine. Time-lapse images were collected at 10-second intervals.
To reveal the dynamic exchange of bodies formed by Oregon Green-EMB1579 and Oregon Green-EMB1579ΔRED, FRAP experiments were performed under an Olympus FV1200MPE laser scanning confocal microscope. For the whole-body FRAP, the bleaching was performed by intense illumination with the 488-nm laser set to 100% and the 405-nm laser set to 40% for 5 seconds. For the half-bleach, the setting was 100% power with 488-nm and 405-nm laser lines for 0.08 seconds. Time-lapse images were collected at 3-second intervals for the wholebody FRAP and 0.2-second intervals for half-FRAP. The rate of fluorescence recovery was measured and calculated with ImageJ using the same method as described above for EMB1579-TGFP FRAP in vivo.
To determine the saturation curve of EMB1579 and EMB1579ΔRED, EMB1579 and EMB1579ΔRED were preclarified under 14,000g for 20 minutes at 4˚C. EMB1579 ranging from 200 to 500 nM and EMB1579ΔRED ranging from 200 to 1,000 nM were incubated in P buffer (25 mM Tris-HCl [pH 8.0], 100-3,000 mM KCl, 2 mM DTT) and centrifuged at 600g, 20˚C for about 3 hours [84]. The supernatant was collected and the protein concentration in the supernatant was measured using the Bradford assay and A280.

Visualization of the recruitment and condensation of proteins by compartments formed by EMB1579 in vitro
EMB1579, mCherry-MSI4, mCherry-DDB1B, and mCherry were preclarified under 14,000g for 20 minutes at 4˚C. To determine whether MSI4 and DDB1B can be recruited into EMB1579 compartments, Oregon Green-labeled EMB1579 (2.5 μM) was incubated with mCherry-DDB1B (500 nM) or mCherry-MSI4 (20 nM) or mCherry (12 μM) in F-buffer for 5 minutes at room temperature in the dark. mCherry-DDB1B, mCherry-MSI4, or mCherry was used as the control under the same conditions. The mixtures were injected into a flow chamber with the coverslip pretreated with Poly-L-lysine and observed under an Olympus FV1200MPE laser scanning confocal microscope with excitation wavelengths set at 488 nm for Oregon Green and 561 nm for mCherry. The partition coefficient was defined as the ratio of the protein in condensates versus that in dilute phase, which was calculated according to the published method [85].
To determine whether EMB1579 condensates can condense MSI4 from Arabidopsis total protein extract, 2-week-old Arabidopsis roots expressing pFGC5941-RFP-MSI4 were collected and ground in buffer (50 mM Tris-HCl [pH 8.0], 2 M KCl, 4 mM DTT, complete EDTA-free Protease Inhibitor Cocktail) on ice. The extract was centrifuged under 14,000g at 4˚C for 20 minutes and the supernatant (containing the protein) was subsequently separated into two equal parts. One part was dialyzed with EMB1579 in buffer P with PEG 3350 for phase separation and the other part was added into the same volume of buffer X and dialyzed in buffer P with PEG 3350 as a control. After 1-hour dialysis, the mixed proteins were injected into a flow chamber with the coverslip pretreated with Poly-L-lysine and observed under an Olympus FV1200MPE laser scanning confocal microscope with excitation wavelength set at 561 nm for RFP.

Pull-down experiments and mass spectrometry analysis
Two-week-old roots of WT Arabidopsis seedlings were collected and soaked in protein extraction buffer (50 mM Tris-HCl [pH 8.0], 50 mM KCl, 2 mM DTT) for about 1 hour. The protein extraction buffer was discarded and the samples were ground on ice with the addition of complete EDTA-free Protease Inhibitor Cocktail. The ground samples were subsequently subjected to centrifugation under 14,000g at 4˚C for 20 minutes. The collected supernatant was dialyzed into P buffer (25 mM Tris-HCl [pH 8.0], 50 mM KCl, 2 mM DTT), and recombinant EMB1579-6×His protein bound to Ni-NTA beads was subsequently added into the supernatant. Ni-NTA beads were used as the control. After incubation for 2 hours on ice, the Ni-NTA beads were briefly washed twice with buffer P. The protein samples were separated by SDS-PAGE and detected by silver staining. The bands of interest were cut out and the proteins were identified by mass spectrometry (OrbiTrap Fusion LUMOS, Thermo).

RNA-seq
Total RNA was isolated from 7-day-old WT and emb1579-3 Arabidopsis seedlings with TRIzol. Seedling cDNA libraries for WT and emb1579-3 were constructed using RNA-seq for pairedend transcriptome sequencing, which was conducted by Shanghai Majorbio Biopharm Technology (Shanghai, China). Poly(A) mRNA was isolated using Sera-mag Magnetic Oligo (dT) Beads and fragmented into short pieces. Double-strand cDNA was obtained with random hexamers and reverse transcriptase. RNA-seq libraries were constructed and sequenced using Illumina HiSeq 4000 in Shanghai Majorbio Bio-pharm Biotechnology (Shanghai, China). The RNA-seq reads were aligned to the TAIR10 Arabidopsis Genome using TopHat v2.0 [86].

ChIP-seq
Chromatin was extracted from 7-day-old Arabidopsis seedlings of WT and emb1579-3 mutants. Anti-H3K27me3 antibodies were used for ChIP. After sonication to break the DNA into pieces, the DNA fragments isolated by ChIP were used to construct DNA pools with a NEBNext Ultra II DNA Library Prep Kit for Illumina following the manufacturer's instructions. After six cycles of PCR, PCR products were confirmed to be in the size range 200-500 bp by DNA agarose gel electrophoresis. The 200-to 500-bp DNA bands were collected and further used for PCR amplification. The PCR products were purified with magnetic beads and used for sequencing, which was carried on an Illumina HiSeq with sequencing by synthesis in Shanghai Romics Biotechnology (Shanghai, China). e.m, n = 3. Numerical data underlying this figure are available in S1 Data. (B) Western blot analysis to determine the relative amount of EMB1579-TGFP and EMB1579ΔRED-TGFP in the nucleus. Total nuclear proteins from Arabidopsis seedlings were probed with anti-GFP antibody. H3 protein (detected with an anti-H3 antibody) was used as the loading control. The original pictures are available in S1 Raw Images. (C) Images of 7-day-old Arabidopsis seedlings growing on plates. Bar = 0.5 cm. (D) Quantification of primary root length of 7-day-old seedlings in WT, emb1579, and its complementation lines. Data are presented as mean ± s.e.m. ��� P < 0.001 by Student t test. Numerical data underlying this panel are available in S1 Data. (E) Images of 6-week-old Arabidopsis plants growing in pots. Bar = 2 cm. (F) Quantification of the number of rosette leaves at bolting in WT, emb1579, and the complementation lines. Data are presented as mean ± s.e.m. ��� P < 0.001 by Student t test. Numerical data underlying this panel are available in S1 Data. EMB1579, EMBRYO DEFECTIVE 1579; ND, no significant difference; qRT-PCR, quantitative reverse transcription PCR; TGFP, tandem copies of enhanced green fluorescent protein; WT, wild type.

S4 Movie. Time-lapse images indicating the fusion of EMB1579 condensates in vitro.
Time-lapse images of EMB1579 compartments were collected every 10 seconds and then compressed into a movie with a display rate of 10 frames per second. The fusion events are indicated by different-colored arrows. Bar = 5 μm. EMB1579, EMBRYO DEFECTIVE 1579. (AVI) S5 Movie. Time-lapse images indicating the recovery of fluorescence of an EMB1579 condensate after half of it was bleached in vitro. Half-bleaching was performed with an in vitroformed EMB1579 condensate. Time-lapse images were captured every 0.2 seconds and then compressed into a movie with a display rate of 15 frames per second. Bar = 1 μm. EMB1579, EMBRYO DEFECTIVE 1579. (AVI) S1  Table. qRT-PCR validation primers. qRT-PCR, quantitative reverse transcription PCR. (DOC) S1 Data. Individual sheets for the underlying numerical data for Figs 1-6 and figures in supporting information.
(XLS) S1 Raw Images. Unprocessed images of all gels and blots in the paper. (PDF)