The H1/H5 domain contributes to OsTRBF2 phase separation and gene repression during rice development

Abstract

Transcription factors (TFs) tightly control plant development by regulating gene expression. The phase separation of TFs plays a vital role in gene regulation. Many plant TFs have the potential to form phase-separated protein condensates; however, little is known about which TFs are regulated by phase separation and how it affects their roles in plant development. Here, we report that the rice (Oryza sativa) single Myb TF TELOMERE REPEAT-BINDING FACTOR 2 (TRBF2) is highly expressed in fast-growing tissues at the seedling stage. TRBF2 is a transcriptional repressor that binds to the transcriptional start site of thousands of genes. Mutation of TRBF2 leads to pleiotropic developmental defects and misexpression of many genes. TRBF2 displays characteristics consistent with phase separation in vivo and forms phase-separated condensates in vitro. The H1/H5 domain of TRBF2 plays a crucial role in phase separation, chromatin targeting, and gene repression. Replacing the H1/H5 domain by a phase-separated intrinsically disordered region from Arabidopsis (Arabidopsis thaliana) AtSERRATE partially recovers the function of TRBF2 in gene repression in vitro and in transgenic plants. We also found that TRBF2 is required for trimethylation of histone H3 Lys27 (H3K27me3) deposition at specific genes and genome wide. Our findings reveal that phase separation of TRBF2 facilitates gene repression in rice development.

An essential protein domain of a transcriptional repressor promotes its phase separation, chromatin targeting, and gene repression during rice development.

Introduction

Compartmentalization is commonly used by cells for the spatial organization of cellular materials and metabolic processes. Recent studies have revealed that the biomolecular condensates driven by liquid–liquid phase separation (LLPS) are essential for forming membraneless compartments (Banani et al. 2017; Emenecker et al. 2021). A few features have been identified in these phase-separated proteins, including intrinsically disordered region (IDR), low-complexity domain (LCD), and prion-like domain (Oldfield and Dunker 2014; Fang et al. 2019; Sprunger and Jackrel 2021), which lack stable structure yet display essential function. Little is known about the contribution of the well-structured domains in LLPS. These phase-separated proteins regulate gene expression, chromatin structure, environmental perception, and so on (Hyman et al. 2014; Boija et al. 2018; Gibson et al. 2019; Emenecker et al. 2021; Chen et al. 2022; Wang et al. 2022).

Transcription is a tightly controlled process in which many proteins must be assembled at the chromatin of the transcribing genes. Transcription factors (TFs) and transcriptional machinery complexes are central to this process. TFs bind to their cognate DNA motifs and recruit cofactors to induce or repress gene expression. The transcriptional complex containing RNA Polymerase II and cofactors moves along the template DNA to transcribe RNA. Increasing evidence has demonstrated that concentrating relevant regulators in the membraneless compartments via LLPS is beneficial to transcriptional regulation (Hyman et al. 2014; Alberti 2017; Hnisz et al. 2017; Peng et al. 2020; Wagh et al. 2021). A computational simulation proposed a model of gene regulation relying on the phase-separated multimolecular assembly of TFs and cofactors, which can explain the features of transcriptional control (Hnisz et al. 2017). The FET protein family protein Ewing sarcoma forms phase-separated protein hubs through the interaction of the LCD domain to stabilize DNA binding and activate transcription in mammals (Chong et al. 2018). The transcription coactivator Mediator Complex Subunit 1 (MED1) contains an IDR region responsible for its phase separation ability in vitro. Transcription activators such as octamer-binding protein 4 (OCT4) and GCN4 can interact with MED1 through their activation domains (ADs), forming phase-separated protein condensates to promote gene activation (Boija et al. 2018). Many plant TFs potentially form phase-separated condensates, implicating critical functions of phase separation on gene transcription (Liu et al. 2006; Salladini et al. 2020). These examples have demonstrated the critical roles of phase separation in gene activation. However, its roles in transcriptional repression are largely unknown.

The Single Myb Histone 1 (SMH1) family proteins were identified in maize (Zea mays) through the sequence similarity to the Myb domain of human (Homo sapiens) telomere protein telomeric repeat-binding factor 1 (TRF1) (Marian et al. 2003). TRF1 binds to double-stranded telomeric repeats through the single C-terminal Myb domain (van Steensel and de Lange 1997). Unlike fungi and animals, the plant SMH1 proteins exhibit 2 unique features. First, the Myb domain is located at the N-terminal. Second, the plant SMH1 proteins contain a central linker-histone H1/H5 domain and a C-terminal coiled-coil (CC) domain (Marian et al. 2003). So, the SMH1 family is a plant-unique protein family. Like other telomeric proteins, ZmSMH1 binds to double-stranded telomeric repeat sequences through the Myb domain (Marian et al. 2003; Byun et al. 2008; Hofr et al. 2009). Meanwhile, ZmSMH1 interacts with other telomeric proteins. The Arabidopsis (Arabidopsis thaliana) SMH1 family proteins TELOMERE REPEAT-BINDING PROTEIN1 (AtTRB1), AtTRB2, and AtTRB3 directly interact with the catalytic subunit of telomerase (TERT) (Schrumpfová et al. 2014). The triple mutant plants exhibit telomere shortening phenotype (Zhou et al. 2018). In rice (Oryza sativa), there are 3 TELOMERE REPEAT-BINDING FACTOR (TRBF) proteins (Byun et al. 2008). Whether rice TRBF proteins directly control telomere length has yet to be determined. A study links TRBF1 to telomere length regulation that the TELOMERE REPEAT-BINDING FACTOR LIKE 1 (OsTRFL1) associates with TRBF1, and the ostrfl1 plants exhibit longer telomeres (Byun et al. 2018). However, TRBF proteins have yet to be carefully characterized in rice.

Recently, a few studies suggested that SMH1 proteins are also involved in transcriptional regulation. In Arabidopsis, the AtTRB2 directly interacts with the histone deacetylases AtHDT4 and AtHDA6 (Lee and Cho 2016), thereby regulating histone acetylation at heterochromatin regions potentially through the RNA-directed DNA methylation pathway (Tan et al. 2018). AtTRBs associate with core Polycomb Repressive Complex 2 (PRC2) components CURLY LEAF (AtCLF) and SWINGER (AtSWN) and recruit PRC2 to the target chromatin (Zhou et al. 2016, 2018). Here, we characterized the function of TRBF2 in rice. TRBF2 is a transcription repressor. Mutation of TRBF2 results in severe developmental defects and derepression of many genes. TRBF2 displays the characteristics of biomolecular condensates driven by LLPS in the nucleus and forms phase-separated condensates in vitro. The H1/H5 domain of TRBF2 is required for its phase separation, efficient chromatin targeting, and gene repression. Replacing the H1/H5 domain with a phase separation domain from AtSERRATE (AtSE) partially recovers the function of TRBF2 in gene repression. We also displayed that, like AtTRBs, TRBF2 promotes Polycomb silencing of specific genes and genome widely in rice. Thus, our data reveal that TRBF2 plays an essential role in gene repression, and LLPS contributes to TRBF2-mediated gene repression in rice development.

Results

TRBF2 regulates rice development

TRBF2 (Os12g0613300) is one of the 3 TRBF proteins in rice (Fig. 1A), which shares high sequence similarity with SMH1 proteins in other plant species (Supplementary Fig. S1). Phylogenetic analysis of SMH1 family proteins revealed that 1 SMH1 protein evolved initially in Streptophyta in Klebsormidiophyceae and then diversified to 3 homologs in mosses and several diverse homologs in angiosperms (Fig. 1A) (Kusová et al. 2023).

Figure 1.

Identification and expression analysis of TRBF2. A) A phylogenetic tree analysis of plant SMH family proteins. Os, O. sativa; Zm, Z. mays; At, A. thaliana; Le, Lycopersicum esculentum; Cs, Cucumis sativus; Atr, Amborella trichopoda; Eg, Elaeis guineensis; Smo, Selaginella moellendorffii; Kn, Klebsormidium nitens; Sp, Sphagnum fallax; Pp, Physcomitrium patens. B) TRBF2 expression levels in various organs revealed by RT-qPCR. Sd, seedling; Rt, root; St, stem; L, leaf; Fl, flag leaf; Sh, sheath; P, panicles; Se, seed. Values are mean ± Sd (n = 3 biological replicates). The OsActin1 gene was used as an internal reference control. C to I) Histochemical detection of ProTRBF2:GUS expression in transgenic rice plants. GUS activity was observed in seed C), 2-d-old seedling D), 5-d-old seedling E), flag leaf F), sheath G), pistil H), and anther I). Scale bars, 2 mm (C, D); 5 mm (E to G); 500 μm (H, I).

Reverse transcription quantitative PCR (RT-qPCR) revealed that TRBF2 is broadly expressed in various organs, with higher expression in the leaf and leaf sheath (Fig. 1B). To better examine the expression pattern, a 2.7-kb TRBF2 promoter was fused to the GUS reporter gene. The construct was transformed into wild-type (WT) rice (O. sativa subsp. japonica) variety Nipponbare (Nip). Most of the ProTRBF2:GUS transgenic lines exhibited similar staining patterns. The GUS signal was detected at the embryo region of the dry seeds (Fig. 1C). The GUS signal was also detected at the tip of the coleoptile, the leaf blade, the tips of the radicle, and the crown roots of the young seedlings (Fig. 1, D and E). We also detected the GUS signal in the flag leaf and leaf sheath (Fig. 1, F and G) and in the male and female reproductive organs such as stamen and pistil (Fig. 1, H and I). In general, TRBF2 is ubiquitously expressed with stronger expression in the fast-growing tissues at the seedling stage, indicating that TRBF2 may play a regulatory role in rice development.

To study the biological function of TRBF2, the CRISPR/Cas9 genome editing method was employed to generate the loss of function mutants in Nip (Ma et al. 2015). Two single guide RNAs (sgRNAs) were designed, one targeting the linker region between the Myb and the H1/H5 domains and the other targeting the H1/H5 domain (Fig. 2A). Three independent mutants, producing premature stop codons, were identified by Sanger sequencing (Fig. 2B and Supplementary Fig. S2A). The mutated mRNA still could be detected in the mutants with lower levels compared with the WT by RT-qPCR (Fig. 2C), which could be caused by reduced mRNA stability. We sequenced the mutated cDNA and confirmed the mutation type of each TRBF2 mutant (Fig. 2B and Supplementary Fig. S2A). All the trbf2 lines displayed obvious developmental defects, such as severe dwarfism at all the developmental stages compared with WT (Fig. 2, D to F). Because of the similar developmental defects of 3 trbf2 mutant lines, except where indicated, trbf2-1 was used for further analysis and referred to as trbf2 in the following study. Rice height is determined by the internode number and the length of each internode at the mature stage. trbf2 has the same internode number as WT, but a reduced length of each internode (Fig. 2, G and H), indicating the short stature of trbf2 was due to shorter internodes. trbf2 also displayed reduced panicle length and decreased numbers of primary and secondary panicle branches compared with WT (Fig. 2, I to K). In addition, the mutants showed narrower kernels and grains compared with WT (Supplementary Fig. S2, B to D). These growth defects were consistent with the higher expression of TRBF2 at the dividing tissues.

Figure 2.

TRBF2 regulates rice development. A) A schematic picture shows the conserved domains of the TRBF2 protein. The positions of the 2 sgRNAs are shown in red lines. B) The sequence of selected sgRNA and corresponding protospacer adjacent motif (PAM) for each target site are displayed in blue. The indels are shown in red nucleotides or minuses. The altered amino acid sequences in the trbf2-1, trbf2-2, and trbf2-3 mutants are highlighted in red. All 3 types of mutations were confirmed by cDNA sequencing of TRBF2 mutants. C) The mRNA levels of TRBF2 in the mutants detected by RT-qPCR. Relative levels to WT are presented. Values are mean ± Sd (n = 3 biological replicates). Asterisks indicate significant differences (**P < 0.01, ***P < 0.001, Student's unpaired 2-tailed t-tests). D to F) Representative plants of trbf2 compared with WT at the seedling stage D), the vegetative growth stage E), and the heading stage F). Scale bars: 10 cm. G to K) Comparison of the agronomic traits between WT and trbf2. Main internode length G, H), main panicle length I), and the number of panicle branches J, K). Values are mean + Sd. The quantified individuals are H) 20 and K) 30. Scale bars: 5 cm.

TRBF2 is a telomere repeat-binding protein (Byun et al. 2008), so we first compared its subcellular localization with other telomere regulators. The RICE TELOMERE-BINDING PROTEIN1 (RTBP1) is colocalized with telomere repeats in vivo and required for maintaining telomere length and proper architecture of telomeres (Hong et al. 2007). OsTRFL1 binds to double-stranded telomeric repeats and interacts with TRBF1 (Byun et al. 2018). Like other TRB proteins, TRBF2 displayed foci in the nucleus when transiently expressed in the Nicotiana benthamiana leaves (Supplementary Fig. S3A) (Zhou et al. 2016; Byun et al. 2018). These TRBF2 foci were mostly colocalized with RTBP1 and OsTRFL1 (Supplementary Fig. S3A), suggesting TRBF2 may regulate telomere function. Second, TRBF2 bound to the telomere repeats was detected by an electrophoretic mobility shift assay (EMSA) (Supplementary Fig. S3B). Deleting the Myb domain (TRBF2ΔMyb) largely disrupted the binding activity compared with WT TRBF2, and the Myb domain alone (TRBF2-Myb) could bind to the telomere repeats (Supplementary Fig. S3B). The H1/H5 domain of AtTRB1 has nonspecific DNA-binding activity (Mozgová et al. 2008). So, we tested whether the H1/H5 domain of TRBF2 contributes to telomere-binding activity. The data displayed that the TRBF2ΔH1/H5 still binds to telomere repeats, and H1/H5 cannot bind to the telomere repeats or with very low capability (Supplementary Fig. S3B). TRBF2 and any truncations tested in our experimental conditions could not bind to a nonspecific double-stranded DNA (Supplementary Fig. S3C). These observations supported that TRBF2 could bind to the telomere repeats in vitro, and the Myb domain plays a major role in the binding (Byun et al. 2008; Mozgová et al. 2008). Third, terminal restriction fragment analysis showed that the trbf2 plants had similar telomere length to WT plants (Supplementary Fig. S3D). AtTRB proteins display functional redundancy in controlling telomere length in Arabidopsis. attrb1 trb2 trb3 and attrb1 trb3 trb2−/+ show shorter telomere compared with the WT Arabidopsis plant. However, the telomere length is not obviously changed in attrb2 compared with WT rice (Zhou et al. 2016, 2018). So, we asked whether the increased expression of other TRBF genes compensates for TRBF2 function. The expression levels of TRBF1 and TRBF3 were not obviously altered in trbf2 measured by RT-qPCR (Supplementary Fig. S3E). These data displayed that TRBF2 has telomere repeat-binding activity. However, the developmental defects of trbf2 are plausible independent of the telomere regulation.

TRBF2 suppresses a set of key developmental genes

TRB proteins can work as TFs together with histone modifiers to repress target gene expression (Zhou et al. 2016, 2018; Tan et al. 2018; Wang et al. 2023). We proposed that TRBF2 may regulate rice development through its TF activity. First, we assayed the transcriptional activity of TRBF2 by a modified dual-luciferase (LUC) assay (Fig. 3A). The firefly LUC gene was used as the reporter; the CaMV 35S promoter (35S)-driven Renilla (REN) LUC gene in the same construct was used as an internal control (Fig. 3A). TRBF2 expression resulted in lower LUC/REN activity than the empty vector (Fig. 3B), displaying a transcriptional repressor activity of TRBF2. Second, we investigated the role of TRBF2 on gene regulation at the genome-wide level. Two-week-old plants of WT and trbf2 without roots were harvested for total RNA extraction. A transcriptome analysis was performed by messenger RNA sequencing (RNA-seq). A total of 1,972 differentially expressed genes (DEGs) were identified, including 1,034 up- and 938 down-DEGs (Fig. 3C; Supplementary Fig. S4, A and B, Table S1, and Data Set S1).

Figure 3.

The trbf2 is defective in gene expression at the genome-wide level. A) Schematics of the constructs used for transcription activity assay. GAL-DB, GAL4 DBD. B) TRBF2 has transcriptional repression activity in rice. The reporters and effectors were coexpressed in rice protoplasts, and REN and LUC activity were measured. The relative LUC activities normalized to the REN activities are shown (LUC/REN). Values are mean ± Sd (n = 3 biological replicates, 2 technical replicates each time). Asterisks indicate significant differences (***P < 0.001, Student's unpaired 2-tailed t-tests). C) Volcano plot showing the DEGs between WT and trbf2 analyzed by RNA-seq. The x axis indicates fold change in expression, and the y axis indicates the magnitude of expression. The size of each dot represents the significance of the P-value. D) Boxplots showing the expression levels of normal (28,910), upregulated (1,034), and downregulated (938) genes analyzed by reads intensity of RNA-seq of WT and trbf2. The lower and upper ± 1.5 quartiles are indicated by whiskers, the lower and upper ends of the boxes indicate the 25th and 75th quartiles, and the line across the middle of the box identifies the median sample value. The P-value is calculated by the paired 2-tailed t-tests. E) Integrative Genomics Viewer visualization showing derepression of Os06g0602400, Os08g0419100, Os11g0230400, and Os11g0532200 in trbf2 compared with WT. The y axis indicates reads density. F) Relative expression levels of Os06g0602400, Os08g0419100, Os11g0230400, and Os11g0532200 in WT, trbf2-1, trbf2-2, and trbf2-3 seedlings measured by RT-qPCR. Relative levels to WT are presented. Values are mean ± Sd (n = 3 biological replicates). Asterisks indicate significant differences (*P < 0.05, **P < 0.01, ***P < 0.001, Student's unpaired 2-tailed t-tests).

Most DEGs overlapped with protein-coding genes (Supplementary Fig. S4C). Gene ontology (GO) analysis indicated that genes involved in response to secondary metabolic processes, system development, cell death, and flower development were overrepresented in the DEGs (Supplementary Fig. S4D). We then compared the expression levels of these trbf2-resulted DEGs with unchanged genes in WT. The downregulated DEGs displayed relatively similar expression levels as unchanged genes, and the upregulated DEGs were not or low expressed in WT (Fig. 3D), consistent with the repression role of TRBF2. We focused our analysis on the up-DEGs. Four genes, which all displayed higher expression levels in trbf2, were selected to verify the sequencing results (Fig. 3E). The expression level of these 4 genes was also derepressed in all 3 TRBF2 mutants (Fig. 3F), supporting that the RNA-seq data were highly reliable.

To understand how TRBF2 regulates plant growth and development, we mapped genome-wide occupancy of TRBF2 using chromatin immunoprecipitation followed by sequencing (ChIP-seq). To this end, transgenic plants carrying ProUBI:TRBF2-Flag-HA/WT were generated and crossed with trbf2. The transgene complemented the developmental defects of trbf2, showing that the transgene functions similarly to the endogenous protein (Fig. 4A). ChIP-seq libraries were prepared in 3 biological replicates using ProUBI:TRBF2-Flag-HA/WT seedlings (Supplementary Fig. S5A and Table S1). In total, 8,807 TRBF2-enriched regions were identified, and more than half of these regions were located at the 5′ end of genes near the transcription start site (TSS) (Fig. 4B). TRBF2 was enriched at 7,347 genes (Supplementary Data Set S2), mostly enriched at the flower, shoot system, and system development processes by GO analysis (Supplementary Fig. S5B). It is consistent with the developmental defects of trbf2 mutants; supporting TRBF2 is critical for rice development. Metagene plots displayed that TRBF2 is almost exclusively enriched at the TSS of genes (Fig. 4C). The motifs of TRBF2-binding peaks were scored according to the degree of enrichment and the probability of being located at the center of enriched regions. The telobox was the most significantly enriched motif (Supplementary Fig. S5C), consistent with the telomeric sequence-binding activity of TRBF family proteins (Byun et al. 2008). We then compared the TRBF2-binding genes with the up-DEGs in trbf2. A total of 194 genes were identified, inducing several key development regulators, such as SMALL GRAIN AND DWARF 2 (SGD2), MADS-BOX TRANSCRIPTION FACTOR 27 (MADS27), and Os10g0372800 (Supplementary Fig. S5D and Data Set S2) (Chen et al. 2018, 2019). These genes are more likely to be directly repressed by TRBF2.

Figure 4.

Analysis of TRBF2 target sites in seedlings. A) Image showing WT, trbf2-1, and 2 lines of ProUBI:TRBF2-Flag-HA/trbf2-1 at the heading stage. Scale bars: 10 cm. B) Regions enriched for TRBF2-binding sites. The proportion of each region was indicated. C) Metagene plots showing TRBF2 intensity at all genes (n = 37,960) and the TRBF2-binding genes (n = 7,347). The y axis indicates reads of the exon model per million mapped reads (CPM, read coverage normalized to 1× sequencing depth). TSS, transcription start site; TES, transcription end site. D) Relative expression levels of MADS27 and Os10g0372800 in WT, trbf2-1, trbf2-2, and trbf2-3 seedlings measured by RT-qPCR. Values are mean ± Sd (n = 3 biological replicates). Relative levels to WT are presented. Asterisks indicate significant differences (***P < 0.001, Student's unpaired 2-tailed t-tests). E) Schematics of the constructs used for dual-LUC assay. F) The transcriptional repression function of TRBF2 on MADS27 and Os10g0372800. The reporters and effectors are coexpressed in rice protoplasts, and REN and LUC activities are measured. The relative LUC activities normalized to the REN activities are shown (LUC/REN). Values are mean ± Sd (n = 3 biological replicates, 2 technical replicates each time). Asterisks indicate significant differences (***P < 0.001, Student's unpaired 2-tailed t-tests).

SGD2, MADS27, and Os10g0372800 were derepressed in trbf2, consistent with the RNA-seq data again (Supplementary Fig. S5, E and F). There are 2 Myb motifs, located at 505- and 635-bp upstream of the translation start codon in the SGD2 promoter, and a telobox motif, located at 1,425-bp upstream of the translation start codon in the MADS27 promoter (Supplementary Fig. S5G). So, we carried out an EMSA to determine whether TRBF2 directly binds to SGD2 and MADS27 promoters. The FAM-labeled probes were designed: P1 contained the 2 Myb motifs in the SGD2 promoter, and P2 harbored the telobox motif in the MADS27 promoter (Supplementary Fig. S5G). P1 or P2 was incubated with purified maltose-binding protein (MBP)-TRBF2 or MBP, and a specific shifted band was observed with MBP-TRBF2, but not MBP alone (Supplementary Fig. S5, H and I). The unlabeled probes could compete with the labeled probes (Supplementary Fig. S5, H and I), suggesting a specific binding activity of TRBF2. Deletions of the Myb motifs of the P1 (P1ΔMyb) or the telobox motif of the P2 (P2Δtelobox) greatly reduced the binding levels of TRBF2 (Supplementary Fig. S5, H and I), suggesting a crucial role of the TRBF2-binding motifs. These results displayed that TRBF2 directly binds the promoters of SGD2 and MADS27 in vitro. Next, we used MADS27 and Os10g0372800 as examples to address the function of TRBF2 on transcription. MADS27 and Os10g0372800 were derepressed in all 3 trbf2 mutants (Fig. 4D). Then, we performed a dual-LUC assay to test whether TRBF2 could repress MADS27 and Os10g0372800 expression directly. The reporter vector contained the LUC reporter driven by the promoter of MADS27 or Os10g0372800; Pro35S:REN in the same construct was used as the internal control (Fig. 4E). TRBF2 repressed MADS27 and Os10g0372800 expression in rice protoplasts (Fig. 4, E and F). Taken together, these results suggest that TRBF2 directly binds to target genes to repress their expression.

TRBF2 forms liquid-like nuclear condensates in vivo and undergoes phase separation in vitro

Previously, we found TRBF2 formed foci in the nucleus in the N. benthamiana leaves (Supplementary Fig. S3A); we examined the subcellular localization of TRBF2 in rice plants. We generated ProTRBF2:TRBF2-Venus/WT transgenic rice, which can complement the mutant phenotype when introduced into trbf2 (Fig. 5A). The lateral root tip was observed under a confocal microscope. The TRBF2-Venus fluorescence was mainly localized in the nucleus with several intense foci (Fig. 5B). AtTRBs and TRBF1 also form nucleus foci, and the AtTRB1 protein is highly mobile (Dvořáčková et al. 2010; Schrumpfová et al. 2014; Byun et al. 2018). These characteristics of TRB proteins reminded us of the condensates formed by transcriptional regulators (Boija et al. 2018; Lu et al. 2018; Zhu et al. 2021). This prompted us to test whether TRBF2-Venus foci have liquid-like characteristics. So, we evaluated the dynamicity of TRBF2-Venus foci by fluorescence recovery after photobleaching (FRAP). The photobleached speckles of TRBF2-Venus could be recovered rapidly in the ProTRBF2:TRBF2-Venus/WT plants (Fig. 5, C and D and Video 1), suggesting that TRBF2 redistributes from the unbleached area to the bleached area and forms nuclear speckles again. We also generated Pro35S:TRBF2-Venus/WT transgenic plants; the transgene could fully complement the mutant developmental defects when introduced into trbf2 by genetic crossing (Supplementary Fig. S6A). Although TRBF2 was highly expressed in the Pro35S:TRBF2-Venus/WT transgenic plants, the TRBF2-Venus speckles were also observed in the nucleus. These speckles showed similar dynamics as in ProTRBF2:TRBF2-Venus/WT in the FRAP assay (Supplementary Fig. S6, B to D and Video 2), suggesting the expression level of TRBF2-Venus may not be crucial for its behavior. These observations displayed that TRBF2 forms biomolecular condensates at the nucleus with liquid-like characteristics.

Figure 5.

TRBF2 forms phase-separated condensates in vivo and in vitro. A) Image showing WT, trbf2-1, and 2 lines of ProTRBF2:TRBF2-Venus/trbf2-1 at the heading stage. Scale bars: 10 cm. B) Fluorescence microscopy of rice lateral root tip cells of ProTRBF2:TRBF2-Venus/WT. The 12-d-old seedlings were used. Scale bars: 5 μm; images are representative of 3 independent experiments. C) FRAP of TRBF2-Venus in rice lateral root tip cells carrying the ProTRBF2:TRBF2-Venus/WT transgene. Time 0 indicates the time of the photobleaching pulse. The white arrow indicates the nuclear body that is bleached. The red arrow indicates the unbleached nuclear body used as the control. Pictures are representative of 6 independent experiments. Scale bars: 5 μm. D) FRAP recovery curves of TRBF2-Venus (green line). The gray line displays the fluorescence variation of unbleached nuclear bodies. Values are mean ± Sd (n = 6). E) Images showing His-TRBF2 forms phase-separated droplets at the concentration of 5 μm. NaCl concentration is 50 mm. Scale bars: 5 μm. Images are representative of 3 independent experiments. F) Images showing that CFP-TRBF2 forms phase-separated droplets at the concentration of 5 μm. NaCl concentration is 50 mm. Images are representative of 3 independent experiments. Scale bars: 5 μm. G) FRAP of CFP-TRBF2 droplets in vitro. Time 0 indicates the time of the photobleaching pulse. CFP-TRBF2 protein is 10 μm, and NaCl concentration is 150 mm. Images are representative of 10 independent experiments. Scale bars: 2 μm. H) FRAP recovery curves of CFP-TRBF2 (green line). The gray line displays the fluorescence variation of unbleached droplets. Values are mean ± Sd (n = 10). I) Fusion of CFP-TRBF2 droplets at 10 μm protein and 50 mm NaCl conditions. Scale bars: 2 μm. Images are representative of 3 independent experiments.

To determine whether TRBF2 can drive phase separation in vitro, the His-tagged recombinant TRBF2 was purified (Supplementary Fig. S7, A and B). TRBF2 protein was eluted by the elution buffer containing 500 mm NaCl. We replaced the elution buffer and diluted the NaCl to a physiologically comparable concentration (50 mm), and spherical droplets of TRBF2 were observed under the microscope (Fig. 5E), displaying that His-TRBF2 forms phase-separated condensates in vitro. To facilitate further characterization of TRBF2 condensates, a fluorescent protein CFP was used to label TRBF2 (Supplementary Fig. S7, A and B). At 50 mm NaCl conditions, the CFP-TRBF2 protein solution became cloudy with the increasing concentration of CFP-TRBF2 (Supplementary Fig. S7C). In contrast, the CFP solution remained clear (Supplementary Fig. S7C). Spherical droplets of CFP-TRBF2, but not the CFP alone, were observed under the microscope (Fig. 5F), supporting that TRBF2 but not the CFP undergoes phase separation. The formation of CFP-TRBF2 droplets could be observed at a very low concentration (0.25 μm) (Supplementary Fig. S7D); the number and size of the droplets increased with increasing CFP-TRBF2 concentration and declined with the increasing concentration of NaCl (Supplementary Fig. S7D). FRAP assay showed that the CFP-TRBF2 signal within droplets recovered shortly after photobleaching, demonstrating that CFP-TRBF2 molecules diffused rapidly within droplets (Fig. 5, G and H and Video 3). Approaching droplets could fuse into big puncta upon contact (Fig. 5I and Video 4). These data collectively displayed that TRBF2 exhibits characteristics consistent with phase-separated protein in vivo and undergoes LLPS in vitro.

As the AtTRB proteins display very similar characteristics as TRBF2 (Dvořáčková et al. 2010; Schrumpfová et al. 2014; Byun et al. 2018), next, we tested whether the Arabidopsis TRBs can display phase-separated characteristics. AtTRB1, the homologous of TRBF2 in Arabidopsis, was used in the following assays (Byun et al. 2008). First, AtTRB1 was expressed in N. benthamiana leaves. Several AtTRB1-Venus foci were observed (Supplementary Fig. S8A), consistent with the previous report (Zhou et al. 2016). We performed a FRAP assay and found that after photobleaching, AtTRB1 redistributed rapidly from the unbleached area to the bleached area (Supplementary Fig. S8, A and B and Video 5). The purified CFP-AtTRB1 formed spherical droplets in vitro as well (Supplementary Fig. S8, C and D). These data displayed that the phase separation characteristics of TRB proteins are probably conserved in Arabidopsis and rice.

The H1/H5 domain contributes to TRBF2 phase separation and transcriptional repression

TRBF2 contains an N-terminal Myb domain, a globular H1/H5 domain in the middle, and a C-terminal CC domain (Fig. 6A). To identify the region that is responsible for TRBF2 phase separation, we first compared the amino acid sequence with human and Arabidopsis linker histone H1, as they contain the H1/H5 domain and form protein condensates (Turner et al. 2018; Gibson et al. 2019; Willcockson et al. 2021; Yusufova et al. 2021; He et al. 2024). However, we did not find any region in TRBF2 that is similar to the disordered domains of human H1 or Arabidopsis H1, and the H1/H5 domains of TRBF2 and AtH1 share a higher similarity (Supplementary Fig. S9A). Second, the disordered domain of TRBF2 was predicted by PONDR-FIT (Xue et al. 2010). Two IDRs were predicted, which cover a large part of TRBF2 (Fig. 6A). The IDR1 contained the C-terminal of the Myb domain, the linker sequence between Myb and the H1/H5 and the N-terminal of H1/H5; the IDR2 included the C-terminal of H1/H5, the linker sequence between the H1/H5 and the CC and the whole CC domain (Fig. 6A). TRBF2 structure was also predicted by AlphaFold2 (Jumper et al. 2021), and these 2 IDRs covered the well-structured domains (Supplementary Fig. S9B). Thus, there was no clear clue about the domain responsible for TRBF2 phase separation.

Figure 6.

The H1/H5 domain of TRBF2 is required for phase separation and gene repression. A) Protein domain architectures of TRBF2 and prediction of IDRs by PONDR (www.pondr.com). B) The number and size of droplets in (Supplementary Fig. S10C) are quantified based on 3 fields of size 124.68 × 124.68 μm². The numbers at the top of each column represent the number of droplets. C) Comparing transcriptional repression function of TRBF2 and truncations on MADS27 and Os10g0372800. The reporters and effectors are coexpressed in rice protoplasts, and REN and LUC activities are measured. The relative LUC activities normalized to the REN activities are shown (LUC/REN). Values are mean ± Sd (n ≥ 4 biological replicates, 2 technical replicates each time). Different letters indicate significant differences tested using 1-way ANOVA/Tukey's multiple range test (P < 0.05). D) Fluorescence microscopy of rice lateral root tip cells that express TRBF2-Venus, TRBF2ΔH1/H5-Venus, or TRBF2-IDR^AtSE-Venus. The 7-d-old seedlings of Pro35S:TRBF2-Venus/trbf2-3, Pro35S:TRBF2ΔH1/H5-Venus/trbf2-3, or Pro35S:TRBF2-IDR^AtSE-Venus/trbf2-3 were used in the analysis. Scale bars: 5 μm. Images are representative of 3 independent experiments. E) Image showing WT, trbf2, 2 lines of Pro35S:TRBF2-Venus/trbf2-3, Pro35S:TRBF2ΔH1/H5-Venus/trbf2-3, and Pro35S:TRBF2-IDR^AtSE-Venus/trbf2-3 at the heading stage. Scale bars: 10 cm. F) Image showing WT, trbf2, 2 lines of ProUBI:TRBF2-Flag-HA/trbf2-3, ProUBI:TRBF2ΔH1/H5-Flag-HA/trbf2-3, and ProUBI:TRBF2-IDR^AtSE-Flag-HA/trbf2-3 at the heading stage. Scale bars: 10 cm. G) Relative expression levels of Os10g0372800, Os08g0419100, and Os06g0602400 in WT, trbf2-3, 2 lines of ProUBI:TRBF2-Flag-HA/trbf2-3, ProUBI:TRBF2ΔH1/H5-Flag-HA/trbf2-3, and ProUBI:TRBF2-IDR^AtSE-Flag-HA/trbf2-3 seedlings measured by RT-qPCR. Values are mean ± Sd (n = 3 biological replicates). Relative levels to WT are presented. Different letters indicate significant differences tested using 1-way ANOVA/Tukey's multiple range test (P < 0.05).

According to the combinations of the predicted IDRs and the domains of TRBF2, we then made several truncations to identify the regions that contribute to TRBF2 phase separation. We used the in vitro phase separation assay and a rice transient expression system to assess the phase separation ability of these truncated proteins. Compared with WT TRBF2, deleting the Myb domain (ΔMyb) considerably reduced the number and size of the phase-separated droplets in vitro, and only 1 big speckle was observed in rice protoplast (Fig. 6B and Supplementary Fig. S10, A to D); ΔCC formed smaller phase-separated droplets in vitro and less condensed foci in rice protoplast (Fig. 6B and Supplementary Fig. S10, A to D); ΔH1/H5 almost lost the phase separation ability in vitro and rice protoplast (Fig. 6B and Supplementary Fig. S10, A to D). However, the H1/H5 domain alone did not form phase-separated protein droplets (Fig. 6B and Supplementary Fig. S10, A to D). These observations suggested that all the Myb, H1/H5, and CC domains contribute to TRBF2 phase separation ability; the H1/H5 domain played an important but insufficient role in triggering TRBF2 phase separation.

As the H1/H5 domain has 2 regions that were predicted as part of IDRs, we next assessed the contribution of the IDR sequences in the H1/H5 domain in condensate formation and compared with WT TRBF2. Deleting the N-terminal of H1/H5 that is predicted as part of the IDR1 (TRBF2ΔH1/H5-N) reduced the size of the condensate in vitro and rice protoplast (Fig. 6B and Supplementary Fig. S10, A to D). TRBF2ΔH1/H5-C displayed a more obvious reduction. Lacking both ends (TRBF2ΔH1/H5-N C) almost abolished condensate formation. TRBF2ΔH1/H5-M (deleting the middle region outside of the IDRs in H1/H5) formed condensates like TRBF2 (Fig. 6B and Supplementary Fig. S10, A to D). Again, these supported a critical role of the H1/H5 domain in TRBF2 condensate formation. The full length of IDR1 together with H1/H5 (TRBF2-IDR1-H1/H5) could form condensates with much-reduced number and size in vitro; however, we could not observe any condensate in rice protoplast (Fig. 6B and Supplementary Fig. S10, A to D). TRBF2-H1/H5-IDR2 could not form protein condensates. Myb-H1/H5 could form condensates in vitro and rice protoplast (Fig. 6B and Supplementary Fig. S10, A to D). Although we did not find a domain responsible for TRBF2 phase separation, our analysis pointed out that the H1/H5 domain plays an essential role in TRBF2 LLPS.

Condensation of transcriptional regulators by the LLPS mechanism facilitates gene regulation (Hyman et al. 2014; Alberti 2017; Hnisz et al. 2017; Chong et al. 2018; Peng et al. 2020; Wagh et al. 2021). So, we tried to address the function of phase separation in TRBF2-mediated gene repression by the dual-LUC assay in rice protoplasts. We focused our analysis on the H1/H5 domain as its essential role in phase separation. TRBF2ΔH1/H5 and TRBF2ΔH1/H5-N C reduced the repression; TRBF2ΔH1/H5-M could repress MADS27 and Os10g0372800 expression similar as TRBF2 (Fig. 6C and Supplementary Fig. S11). The repression activities of these truncations corresponded very well with the phase separation ability. AtSE is a zinc-finger protein, which forms protein condensates through the IDR to control microRNA biogenesis in Arabidopsis (Xie et al. 2021). We next swapped the H1/H5 of TRBF2 by the IDR of AtSE (TRBF2-IDR^AtSE) (Supplementary Fig. S10, A and B). As expected, TRBF2-IDR^AtSE formed protein condensates in vitro and in rice protoplasts similar to TRBF2 (Supplementary Fig. S10, C and D). TRBF2-IDR^AtSE also repressed the expression of MADS27 and Os10g0372800 at a similar level as TRBF2 and TRBF2ΔH1/H5-M (Fig. 6C). Collectively, these data displayed an essential role of TRBF2 phase separation, which is mainly contributed by the H1/H5 domain, in TRBF2-mediated gene repression.

Our analysis further supported the functional importance of TRBF2 phase separation in plants. We made Pro35S:TRBF2-Venus/trbf2-3, Pro35S:TRBF2ΔH1/H5-Venus/trbf2-3, and Pro35S:TRBF2-IDR^AtSE-Venus/trbf2-3 transgenic plants. These transgenes displayed variable but comparable expression levels in different plants (Supplementary Fig. S12A). Supporting our previous analysis, nucleus foci of TRBF2-Venus and TRBF2-IDR^AtSE-Venus were observed in the lateral root tips of corresponding transgenic plants (Fig. 6D). It was hard to observe any nucleus foci in Pro35S:TRBF2ΔH1/H5-Venus/trbf2-3 plants (Fig. 6D). Pro35S:TRBF2-Venus/trbf2-3 transgenic plants displayed a WT-like phenotype; Pro35S:TRBF2ΔH1/H5-Venus/trbf2-3 transgenic plants exhibited a phenotype close to trbf2-3; Pro35S:TRBF2-IDR^AtSE-Venus/trbf2-3 partially complemented trbf2-3 developmental defects (Fig. 6E). Similar phonotypes were also observed in ProUBI:TRBF2-Flag-HA/trbf2-3, ProUBI:TRBF2ΔH1/H5-Flag-HA/trbf2-3, and ProUBI:TRBF2-IDR^AtSE-Flag-HA/trbf2-3 transgenic plants (Fig. 6F and Supplementary Fig. S12B). We measured the expression levels of 3 previously identified genes Os08g0419100, Os06g0602400 (Fig. 3, E and F), and Os10g0372800 (Fig. 4D), which were derepressed in trbf2-3 (Fig. 6G). Their expression levels were restored in ProUBI:TRBF2-Flag-HA/trbf2-3 transgenic plants, but not the ProUBI:TRBF2ΔH1/H5-Flag-HA/trbf2-3 transgenic plants (Fig. 6G). ProUBI:TRBF2-IDR^AtSE-Venus/trbf2-3 could partially restore their expression levels (Fig. 6G). These data suggested that the H1/H5 domain is essential for TRBF2 function, and the IDR from AtSE could partially replace its function in phase separation as well as gene repression in protoplasts and transgenic plants.

The H1/H5 domain facilitates TRBF2 chromatin targeting

We analyzed the function of the H1/H5 domain on TRBF2 chromatin targeting by comparing the genome-wide occupancy of TRBF2ΔH1/H5, TRBF2-IDR^AtSE with TRBF2 using ChIP-seq. Libraries were prepared and sequenced using ProUBI:TRBF2ΔH1/H5-Flag-HA/trbf2-3 #1 and ProUBI:TRBF2-IDR^AtSE-Flag-HA/trbf2-3 #1 (Supplementary Fig. S12C and Table S1). A total of 1,589 TRBF2ΔH1/H5 and 2,948 TRBF2-IDR^AtSE-binding sites were identified, respectively (Supplementary Data Sets S3 and S4), which were far less than the number of TRBF2-binding sites. Compared with WT TRBF2, more proportion of TRBF2ΔH1/H5- and TRBF2-IDR^AtSE-binding sites were located at the proximal region of the promoter (≤1 kb) (Supplementary Fig. S12D). Then, we compared TRBF2-, TRBF2ΔH1/H5-, and TRBF2-IDR^AtSE-binding levels at TRBF2 target sites. The metagene plots and the heatmaps displayed the highest binding level of TRBF2, the intermediate level of TRBF2-IDR^AtSE, and the lowest binding level of TRBF2ΔH1/H5 (Fig. 7A). We then submitted these binding sites to the MEME-ChIP Web tool to predict the enriched motifs (Machanick and Bailey 2011). Consistent with TRBF2, the telobox was the most significantly enriched motif (Supplementary Fig. S12, E and F), this may be related to the telomeric sequence-binding activity of the Myb domain. GT-1 and Spz1 motifs were also enriched in all 3 samples, supporting a large proportion of the overlapped binding sites (Supplementary Fig. S12, E and F).

Figure 7.

The H1/H5 domain contributes to TRBF2 chromatin targeting. A) Metagene plots and heatmaps showing TRBF2, TRBF2ΔH1/H5, and TRBF2-IDR^AtSE enrichment at the TRBF2-binding sites. Peaks aligned at the peak summit of TRBF2 are plotted with 5-kb upstream and downstream regions. The y axis of metagene plots indicates reads of the exon model per million mapped reads (CPM, read coverage normalized to 1× sequencing depth). RP10M, reads per 10 million mapped. B) Venn diagram showing the overlaps of the binding genes among TRBF2, TRBF2ΔH1/H5, and TRBF2-IDR^AtSE. The P-values are calculated by hypergeometric tests. C) Comparing enrichment levels at selected gene sets by metagene plots. The y axis indicates reads of the exon model per million mapped reads (CPM, read coverage normalized to 1× sequencing depth). TSS, transcription start site; TES, transcription end site. D) Schematic diagrams of the SGD2, MADS27, and Os10g0372800 gene structure. UTRs, exons, and introns are indicated by gray boxes, black boxes, and black lines, respectively. E) Enrichment levels of TRBF2 at SGD2, MADS27, and Os10g0372800 assayed in WT, ProUBI:TRBF2-Flag-HA, ProUBI:TRBF2ΔH1/H5-Flag-HA/trbf2-3, and ProUBI:TRBF2-IDR^AtSE-Flag-HA/trbf2-3. Values are mean ± Sem (n = 3 biological replicates). Relative levels to WT are presented. Each examined region, a or b, is shown below each locus in D) (*P < 0.05, **P < 0.01, ***P < 0.001, n.s., no significant, Student's unpaired 2-tailed t-tests).

We mapped these binding sites to genes and found 1,553 TRBF2ΔH1/H5 and 2,855 TRBF2-IDR^AtSE-binding genes, respectively (Fig. 7B and Supplementary Data Sets S3 and S4). 88.1% of TRBF2ΔH1/H5-binding genes (1,368 out of 1,553) and 77.4% of TRBF2-IDR^AtSE-binding genes (2,211 out of 2,855) were bound by TRBF2, respectively (Fig. 7B). On the common target genes of TRBF2 and TRBF2ΔH1/H5 or TRBF2 and TRBF2-IDR^AtSE, TRBF2 had higher mean enrichment (Fig. 7C). We identified 1,229 common binding genes among TRBF2, TRBF2ΔH1/H5, and TRBF2-IDR^AtSE (Fig. 7B). The binding levels of TRBF2ΔH1/H5 and TRBF2-IDR^AtSE were lower than TRBF2; TRBF2-IDR^AtSE had a higher binding level than TRBF2ΔH1/H5 at these common target genes (Fig. 7C). We then compared the enrichments of TRBF2, TRBF2ΔH1/H5, and TRBF2-IDR^AtSE at TRBF2 target genes SGD2, MADS27, and Os10g0372800 by ChIP-coupled qPCR (ChIP-qPCR), and the DNA methyltransferase gene CHROMOMETHYLASE 3a (CMT3a) was used as a negative control (Cheng et al. 2015). The results showed that TRBF2 could be enriched at the chromatin of these genes (Fig. 7, D and E). TRBF2ΔH1/H5 and TRBF2-IDR^AtSE enrichments were reduced at these genes (Fig. 7E). TRBF2, TRBF2ΔH1/H5, and TRBF2-IDR^AtSE did not target CMT3a. These analyses showed that the H1/H5 domain is essential for TRBF2 efficient chromatin targeting. IDR^AtSE partially complements its function in binding strength and sites, which is probably tightly linked to the feature of phase separation.

TRBF2 promotes Polycomb-mediated repression in rice

Previous studies in Arabidopsis displayed that the TRB proteins associate with PRC2 catalytic subunits AtCLF and AtSWN to repress gene expression (Zhou et al. 2016, 2018). TRBF2 also interacted with rice CLF in yeast 2-hybrid (Y2H) assay (Fig. 8A). To roughly map the interaction surface between TRBF2 and CLF, truncation analysis was conducted via Y2H. Both the H1/H5 and CC domains of TRBF2 are required to interact with CLF in yeast cells (Supplementary Fig. S13). The C-terminus of CLF, including the CXC and SET domains, is required for interacting with TRBF2 in yeast cells; deleting either domain disrupts the interaction (Supplementary Fig. S13). The H1/H5 and CC domains of TRBF2 and the CXC and SET domains of CLF are sufficient for TRBF2-CLF interaction (Supplementary Fig. S13B). These data displayed that TRBF2 interacts with CLF through their C-terminal domains in yeast cells.

Figure 8.

TRBF2 promotes Polycomb repression. A) TRBF2 interacts with CLF in yeast cells. The pairs of indicated plasmids were cotransformed and grown on selective solid media lacking leucine (L), tryptophan (W), and histidine (H) (SD-LWH). AD-T and BD-Lam were negative controls; AD-T and BD-P53 were positive controls. Representative pictures from 3 independent experiments are shown. B) CFP-TRBF2 condensates attract YFP-CLF in vitro. Images are representative of 3 independent experiments. Scale bars: 10 μm (all images are on the same scale). C) Images showing CLF-YFP forms aggregates with TRBF2-CFP condensates in nuclei when coexpressed in rice protoplasts. Top panel, expression of CLF-YFP; bottom panel, coexpression of TRBF2-CFP and CLF-YFP. Scale bars: 5 μm. Images are representative of 3 independent experiments. D) Images showing colocalization of TRBF2-CFP (cyan line) and CLF-YFP (yellow line). The fluorescence intensities (y axis) are plotted along the white line in C) (x axis). Images are representative of 3 independent experiments. E) MA plot showing the changes of H3K27me3 reads intensity between WT and trbf2 analyzed by Cut&tag. The x axis indicates the enrichment intensity of each peak, and the y axis indicates log₂(difference in H3K27me3 intensity). F) Heatmaps of H3K27me3 Cut&tag reads density at regions that either gained or lost H3K27me3 in trbf2 compared with WT. Peaks aligned at the peak summit of H3K27me3 are plotted with 5-kb upstream and downstream regions. RP10M, reads per 10 million mapped. G, H) Metagene plots showing H3K27me3 intensity at all H3K27me3-enriched regions (the H3K27me3 enriched regions in WT) G) or all H3K27me3 marked genes (the H3K27me3 marked genes in WT) H). The y axis indicates reads of the exon model per million mapped reads (CPM, read coverage normalized to 1× sequencing depth). Peaks aligned at the peak summit of H3K27me3 are plotted with 5-kb upstream and downstream regions. I) Enrichment of trbf2 upregulated and downregulated genes per H3K27me3 category. Each category is compared with the expected random overlap. The x axis represents the enrichment score. The P-values are calculated with hypergeometric tests. J) Levels of H3K27me3 at SGD2, MADS27, and Os10g0372800 assayed by ChIP-qPCR. Values are mean ± Sem (n = 3 biological replicates). Each examined region, a or b, is shown below each locus in (Fig. 7D) (*P < 0.05, **P < 0.01, n.s., no significant, Student's unpaired 2-tailed t-tests).

Phase-separated protein condensates attract interacting proteins to facilitate gene regulation (Fang et al. 2019; Seif et al. 2020; Zhang, Fan, et al. 2022). For example, the Pc component of Drosophila PRC1 could be enriched by the Mini-Ph—chromatin condensates to stimulate Polycomb silencing (Seif et al. 2020). So, we asked whether TRBF2 condensates influence the PRC2 dynamic. We purified YFP-tagged CLF (YFP-CLF) and found it could not form protein condensate alone (Fig. 8B). CFP-TRBF2 could not induce solo YFP to form condensate (Fig. 8B). However, YFP-CLF puncta were observed when mixed with CFP-TRBF2 and colocalized with CFP-TRBF2 condensates (Fig. 8B). To study the progression of YFP-CLF puncta formation, YFP-CLF and CFP-TRBF2 were mixed under the confocal microscope, and images were taken immediately. YFP-CLF could be attracted and concentrated by CFP-TRBF2 condensates, and a reduced YFP-CLF signal outside of CFP-TRBF2 droplets was also observed during time progression (Video 6). To study the impact of TRBF2 condensates on PRC2 behavior in planta, CLF was expressed alone or coexpressed with TRBF2 in rice protoplasts. Individually expressed CLF-YFP displayed even distribution in nuclei (Fig. 8C). However, several CLF-YFP nuclear bodies were observed once coexpressed with TRBF2-CFP (Fig. 8C). These nuclear bodies were colocalized with TRBF2-CFP condensates (Fig. 8, C and D). These studies suggested that CLF, the catalytic subunit of PRC2, could be attracted by TRBF2 protein droplets to form protein condensates together with TRBF2.

As TRBF2-IDR^AtSE partially restored the developmental defects of trbf2-3, we investigated whether the TRBF2-IDR^AtSE attracts CLF. First, we tested whether TRBF2-IDR^AtSE interacts with CLF in yeast cells. Consistent with our previous data, the H1/H5 domain is required for TRBF2-CLF interaction; replacing this domain with the IDR of AtSE (TRBF2-IDR^AtSE) disrupted the interaction in yeast cells (Supplementary Fig. S14A). The crowded conditions of phase-separated protein provide multiple weak interaction sites for interactors (Banani et al. 2017; Larson and Narlikar 2018). The condensates of TRBF2-IDR^AtSE may provide such conditions for the TRBF2-CLF interaction. We mixed the purified CFP-TRBF2-IDR^AtSE and YFP-CLF, and YFP-CLF was concentrated in the CFP-TRBF2-IDR^AtSE condensates (Supplementary Fig. S14B). Similar results were observed when TRBF2-IDR^AtSE-CFP and CLF-YFP were transiently expressed in rice protoplasts (Supplementary Fig. S14C). The colocalization signal between TRBF2-IDR^AtSE-CFP and CLF-YFP seemed weaker than that of TRBF2-CFP and CLF-YFP (Supplementary Fig. S14, C and D). We also detected less CLF-HA was coprecipitated with TRBF2-IDR^AtSE-Venus than TRBF2-Venus when transiently coexpressed in rice protoplasts (Supplementary Fig. S14E). Meanwhile, TRBF2ΔH1/H5 could not form protein condensates (Supplementary Fig. S10) and did not interact with CLF in yeast cells (Supplementary Fig. S14A). We hardly observed any TRBF2ΔH1/H5 and CLF protein aggregates (Supplementary Fig. S14, B to D). Little CLF-HA was coprecipitated with TRBF2ΔH1/H5-Venus (Supplementary Fig. S14E). These data further supported that the H1/H5 domain contributes to the phase separation ability of TRBF2 and that the phase separation ability of TRBF2 facilitates the TRBF2-CLF interaction.

We further tested whether TRBF2 regulates gene expression by controlling PRC2 function. We measured genome-wide H3K27me3 distribution in trbf2 by Cut&tag (Kaya-Okur et al. 2019) (Supplementary Table S1). In total, 12,378 H3K27me3 enriched regions were identified in trbf2. A total of 2,846 changed H3K27me3 peaks including 2,286 reduced (lost H3K27me3 in trbf2), and 560 elevated peaks (gained H3K27me3 in trbf2) were detected (Fig. 8E and Supplementary Data Set S5). More than 80% of the changed H3K27me3 peaks displayed reduced levels of H3K27me3 in trbf2 (Fig. 8E). Then, the H3K27me3 peaks were aligned according to the center of the peaks (Fig. 8F). A lower mean level of all H3K27me3 peaks was observed in trbf2 compared with WT (Fig. 8G), suggesting TRBF2 is required for H3K27me3 deposition at the genome-wide level. Then, the H3K27me3 peaks were mapped to genes, a lower mean level of all genes, 422 gained and 2,083 lost H3K27me3 genes were detected (Supplementary Data Set S5).

We then compared the H3K27me3 Cut&tag data with the RNA-seq data, the up-DEGs were significantly enriched in genes that lost H3K27me3, and the down-DEGs were significantly enriched in genes that gained H3K27me3 (Fig. 8I). Our data displayed that TRBF2 is required for genome-wide H3K27me3 distribution. SGD2, MADS27, and Os10g0372800 were derepressed in trbf2 and occupied by TRBF2, so we measured the H3K27me3 levels by ChIP-qPCR. These genes were marked by H3K27me3; in addition, H3K27me3 levels were reduced in trbf2 (Fig. 8J), consistent with their derepressed expression levels in trbf2. This suggested that TRBF2 also plays a role in Polycomb repression; the telomere repeat-binding proteins contribute to Polycomb silencing and could be conserved in monocot and dicot plants.

Discussion

During development, the expression of key developmental genes must be precisely controlled. The TFs play a central role in ensuring the proper gene expression in plant development. Recent studies have demonstrated that concentrating transcriptional regulators, such as TFs, mediator complex, RNA Polymerase II, and so on, in the membraneless compartments through the LLPS mechanism contributes to the precise regulation of transcription (Boija et al. 2018; Chong et al. 2018; Sabari et al. 2018). Several transcriptional activators have been identified to form nuclear condensates by themselves or with other factors to promote transcriptional activation (Boija et al. 2018; Chong et al. 2018). In this study, we have shown that a rice single Myb TF, TRBF2, acts as a transcriptional repressor in controlling rice development. The H1/H5 domain of TRBF2 promotes its phase separation, chromatin targeting, and gene repression. These functions tightly link to the phase separation ability of TRBF2. Thus, our study revealed that the LLPS mechanism contributes to TRBF2-driven gene repression in rice.

The TRB proteins were identified through their potential binding ability on the telomere repeats. They interact with each other through the C-terminal H1/H5 and CC domains, suggesting a coordinated function (Byun et al. 2008; Zhou et al. 2016, 2018). As expected, 3 TRBs in Arabidopsis are functionally redundant; only the triple mutant displays shorter telomere and severe developmental defects (Zhou et al. 2016, 2018). All TRBs in rice and Arabidopsis bind to the telomere repeats in vitro (Byun et al. 2008; Mozgová et al. 2008; Lee et al. 2012). We found that the single mutant of TRBF2 in rice exhibited severe growth defects and established TRBF2 functioning as a transcriptional repressor in regulating rice development (Figs. 2 and 3). It suggests that TRBF2 potentially has evolutionarily diverged in rice and mainly functions as a transcriptional repressor in controlling development. As the telomere length was not altered in trbf2 (Supplementary Fig. S3D), TRBF2 transcriptional activity may be independent of its telomere-binding activity. A detailed comparison of all TRBF mutants, even double and triple mutants, is required to distinguish their roles in regulating transcription and telomere length.

TFs bind to their cognate DNA motifs that provide a favorable local chromatin environment to control gene expression (Hyman et al. 2014; Questa et al. 2016; Hnisz et al. 2017; Yang et al. 2017; Boija et al. 2018; Wagh et al. 2021). TFs contain a structurally well-defined DNA-binding domain (DBD) and a separate AD that frequently has IDRs (Soto et al. 2022). The IDRs of ADs can control TF assembly at chromatin by driving phase separation into transcriptional condensates, participating in gene transcription activation (Boija et al. 2018; Han et al. 2020; Peng et al. 2020; Ma et al. 2021). Recently, Trojanowski et al. (2022) found that the multivalent interactions of ADs promote transcriptional activation; the occurrence of phase separation does not significantly enhance this activating effect in a synthetic system. Phase separation of TRBF2 is required for its function in suppressing gene transcription and chromatin targeting (Figs. 6 and 7). Similar to our observation, reducing the phase separation of the MED1-Bromodomain Containing 4 transcriptional coactivator complex also decreases its accumulation at superenhancers (Sabari et al. 2018). Together, these support the idea that the phase separation of transcriptional regulators plays a fundamental role in controlling gene expression. TRBF2 is an example of how the phase separation of a transcriptional repressor regulates its function. Hence, the function of LLPS in gene regulation may show a protein-specific manner.

TRBF2 forms liquid-like condensates in vivo and in vitro (Fig. 5 and Supplementary Fig. S7). All 3 domains of TRBF2 contribute to its phase separation ability, and the H1/H5 domain plays a major role (Fig. 6B and Supplementary Fig. S10). The CC domain of transcriptional activator PDZ-binding motif (TAZ) and the gene AD of OCT4 in mammals are necessary for their phase separation (Boija et al. 2018; Lu et al. 2020). These together suggested a role of the well-structured domains in condensate formation. Deleting the H1/H5 domain disrupts the function of TRBF2; an IDR from AtSE partially complements its function on phase separation, chromatin targeting, and gene repression (Fig. 6 and Supplementary Fig. S10). This supports the role of phase separation in TRBF2-mediated gene repression. However, TRBF2-IDR^AtSE did not fully complement trbf2-3 developmental defects and had fewer target genes than TRBF2 (Figs. 6 and 7). There should be other factors promoting TRBF2 chromatin targeting. The H1/H5 domain of TRB proteins is required for protein–protein interaction (Fig. 8A and Supplementary Fig. S13) (Byun et al. 2008; Mozgová et al. 2008; Schrumpfová et al. 2008). The cofactors functioning with the H1/H5 domain may also regulate TRBF2 chromatin targeting. Our data did not rule out the possibility that the H1/H5 domain may be associated with chromatin, which may strengthen TRBF2–chromatin interaction. Deleting the H1/H5 domain also disrupts these possible activities in TRBF2 chromatin targeting.

TRB proteins are required for genome-wide H3K27me3 distribution in Arabidopsis and rice (Fig. 8, E to H) (Zhou et al. 2016, 2018). There may be 2 ways that TRBF2 promotes PRC2 silencing. First, like in Arabidopsis, TRBF2 binds to cognate DNA motifs to guide PRC2 chromatin targeting (Zhou et al. 2016, 2018). Second, TRBF2 may compact chromatin to stimulate PRC2 activity. The H1/H5 domain-containing protein H1 in mammals and Arabidopsis promotes chromatin condensation, which requires the phase separation ability of H1 (Zhou et al. 2013; Turner et al. 2018; Willcockson et al. 2021; He et al. 2024). The dense chromatin stimulates PRC2 activity (Yuan et al. 2012). Thus, H1 promotes H3K27me3 deposition at a majority of genes (Willcockson et al. 2021; Yusufova et al. 2021; Teano et al. 2023). TRBF2 also contains the H1/H5 domain and forms protein condensates (Fig. 6 and Supplementary Fig. S10). It is plausible to propose a dense chromatin in TRBF2 condensates. More importantly, TRBF2 attracts PRC2 in the condensates (Fig. 8, B to D). This may provide a favorable local chromatin environment for PRC2 to generate a patched pattern of H3K27me3 (Agger et al. 2007; Mansour et al. 2012; Gaydos et al. 2014; Huang and Sun 2022). On the other hand, H1 prevents TRBF2 and PRC2 targeting at telomeres (Teano et al. 2023). However, many TRBF2 condensates are colocalized with telomere repeat (Supplementary Fig. S3A). The chromatin accessibility is key to the interplay of H1 and TRB proteins under these 2 conditions (Teano et al. 2023). It will be intriguing to explore the role of the H1/H5 domain in setting or distinguishing these chromatin states.

The studies in Arabidopsis showed that TRB proteins recruit PRC2 to the target genes by directly interacting with core PRC2 components (Zhou et al. 2016, 2018). TRBF2 interacted with rice CLF (Fig. 8A), suggesting a possibility that TRBF2 may also recruit PRC2 to target chromatin in rice. The phase-separated proteins may serve as a scaffold to assemble cofactors in the condensates (Illingworth et al. 2015; Wani et al. 2016; Lau et al. 2017; Boija et al. 2018; Plys et al. 2019; Tatavosian et al. 2019). For example, EMBRYO DEFECTIVE 1579 (EMB1579) forms nuclear condensates, accumulating a PRC2 component MSI4 in the condensates and promoting Polycomb silencing (Zhang et al. 2020). TRBF2 may similarly attract PRC2 in the TRBF2 condensates to control rice development (Fig. 8, B to D). The phase-separated TRBF2 is the scaffold that attracts and concentrates the client-PRC2 components in the condensates. The crowded conditions of phase-separated TRBF2 may supply multiple interaction surfaces to strengthen the TRBF2-PRC2 interaction. This was supported by that TRBF2-IDR^AtSE did not interact with CLF in yeast cells (Supplementary Fig. S14A). However, TRBF2-IDR^AtSE concentrated CLF in its condensates (Supplementary Fig. S14, B to D). Further characterization of the regulatory role of TRB proteins on Polycomb silencing will provide essential insights into the mechanisms underlying PRC2 targeting, accumulation at target chromatins, and even Polycomb body formation.

Materials and methods

Plant materials and growth conditions

Rice (O. sativa subsp. japonica) variety Nip was used in this study. All the rice materials were grown in either the greenhouse under 14 h of light (approximately 15,000 LX) and 10 h of dark at 28 ℃ or the experimental field of the Wuhan area (30.6, 114.5) in summer. N. benthamiana materials were grown in the greenhouse under 16 h of light (approximately 5,000 LX) and 8 h of dark at 22 ℃.

To generate TRBF2 mutants, 2 guide RNAs targeting TRBF2 were designed (Xie et al. 2017), and the mutants were generated by CRISPR/Cas9 technology (Miao et al. 2013; Ma et al. 2015) and further genotyped. Primers for generating the TRBF2 CRISPR/Cas9 vector are listed in Supplementary Data Set S6.

To generate Pro35S:TRBF2-Venus/WT stable transgenic plants, the coding sequence of TRBF2 was fused with Venus driven by a CaMV 35S promoter into the BamH I and Sal I site of a modified pCAMBIA1300 vector (Xuan et al. 2024). The plasmid was introduced into WT Nip by the Agrobacterium tumefaciens-mediated transformation. The ProUBI:TRBF2-Flag-HA/WT, Pro35S:TRBF2-Venus/trbf2-3, Pro35S:TRBF2ΔH1/H5-Venus/trbf2-3, Pro35S:TRBF2-IDR^AtSE-Venus/trbf2-3, ProUBI:TRBF2-Flag-HA/trbf2-3, ProUBI:TRBF2ΔH1/H5-Flag-HA/trbf2-3, and ProUBI:TRBF2-IDR^AtSE-Flag-HA/trbf2-3 transgenic plants were generated similarly.

To generate ProTRBF2:TRBF2-Venus/WT stable transgenic plants, the full-length genomic fragment of TRBF2, including promoter and terminator, was amplified from the WT plant. The sequence of Venus was inserted before the translational stop codon and then constructed into the Aat II and Sal I site of pCAMBIA1300 vector. The plasmid was introduced into WT Nip by the A. tumefaciens-mediated transformation.

The transgene homozygous lines were selected and used in the experiments. To generate Pro35S:TRBF2-Venus/trbf2-1, ProUBI:TRBF2-Flag-HA/trbf2-1, and ProTRBF2:TRBF2-Venus/trbf2-1 complementation lines, Pro35S:TRBF2-Venus/WT, ProUBI:TRBF2-Flag-HA/WT, and ProTRBF2:TRBF2-Venus/WT homozygous lines were crossed with trbf2-1. The transgene homozygous lines were selected according to resistance selection, and trbf2-1 was selected by genotyping. All primers used are listed in Supplementary Data Set S6.

GUS staining

For tissue-specific expression analysis, a 2,694-bp fragment upstream from the ATG start code of TRBF2 was cloned into the Kpn I and Sbf I site of a modified pCAMBIA1300 vector to generate ProTRBF2:GUS transgenic plants. GUS activity was analyzed using ProTRBF2:GUS T2 transgene homozygous rice plant tissues by histochemical staining assay. Briefly, plant samples were placed in the staining buffer (50 mm PBS pH 7.0, 10 mm EDTA, 0.1% [v/v] Triton X-100, and 0.5 mg/mL X-Gluc). After vacuum infiltration for 30 min, the staining reaction was kept by incubation at 37 °C. Chlorophylls were removed by incubation in a solution of 70% ethanol. All primers used are listed in Supplementary Data Set S6.

Dual-LUC assay

The dual-LUC assay was performed as described (Hellens et al. 2005). The dual-LUC assay system was used for analyzing the transactivation assay. For the reporter, 5 copies of upstream activation sequence (UAS) (GAL4 UAS) were synthesized and cloned into pGreenII 0800-LUC reporter vectors as a transcriptional fusion with the firefly LUC gene, and mini 35S was constructed after UAS. In the same vector, the REN LUC reporter gene driven by the CaMV 35S promoter was used as an internal control. For effectors, yeast transcriptional activator GAL4 DBD (GAL4BD) fused with the TRBF2 was cloned into pCAMBIA1300 vector. The empty plasmid GAL4BD was used as a negative control.

For transient expression regulation assay, the promoter of MADS27 (1,526-bp upstream of the ATG start code) or Os10g0372800 (1,491-bp upstream of the ATG start code) was amplified and cloned into the Hind III and BamH I site of pGreenII 0800-LUC reporter vector as a transcriptional fusion with the firefly LUC gene. The REN LUC reporter gene driven by the CaMV 35S promoter was used as an internal control in the same vector. The coding sequences of TRBF2, ΔH1/H5, ΔH1/H5-N C, ΔH1/H5-M, and IDR^AtAE were fused with Venus driven by the CaMV 35S promoter into the BamH I and Sal I site of pCAMBIA1300 vector. The reporter and effector constructs were coexpressed in rice protoplasts and then incubated overnight at 28 °C in the dark. The cultured cells were harvested, and the LUC activities were measured using Dual-Luciferase Reporter Gene Assay Kit (Yeasen, 11402ES60). The transcriptional activity was determined by the ratio of the activity of LUC to REN. Relevant primers are listed in Supplementary Data Set S6.

Terminal restriction fragment analysis

The terminal restriction fragment experiment was performed as described previously with modifications (Byun et al. 2018). Briefly, 3-µg genomic DNA was extracted from 2-wk-old seedlings. The DNA was then digested by BamH I and EcoR I (NEB) at 37 °C overnight. The digested DNA was electrophoresed on an agarose gel and blotted to a nylon membrane (Roche, 11417240001). Oligonucleotide (TTTAGGG)₇ was 5′ end-labeled with biotin and used as a probe for Southern blotting.

Protein expression and purification from Escherichia coli

Coding sequences of TRBF2 and truncations were amplified and cloned into the BamH I and Sal I site of the His-mCFP vectors or MBP vectors and then transformed into E. coli BL21 (DE3) strain. Precultured cells containing responding plasmids were grown at 37 °C to reach OD₆₀₀ about 0.6. Protein expression was induced by the addition of 0.5 mm isopropyl β-D-1-thiogalactopyranoside. After being grown for 16 h at 16 °C, cells were harvested and resuspended in suspension buffer (His-mCFP vectors: 20 mm Tris-HCl pH 8.0, 500 mm NaCl, 1 mm EDTA, and 1 mm PMSF; MBP vectors: 10 mm HEPES-KOH pH 7.5, 150 mm NaCl, and 1 mm PMSF). The cells were lysed using a high-pressure homogenizer (ATS Engineering) and centrifuged at 16,000 × g for 1 h. The supernatants flowed through a Ni-NTA column (Thermo, 25215). The Ni-NTA beads were washed with wash buffer (His-mCFP vectors: 20 mm Tris-HCl pH 8.0, 500 mm NaCl, 1 mm EDTA, 15 mm imidazole, and 1 mm PMSF; MBP vectors: 10 mm HEPES-KOH pH 7.5, 150 mm NaCl, and 1 mm PMSF); proteins were eluted by elution buffer (His-mCFP vectors: 20 mm Tris-HCl pH 8.0, 500 mm NaCl, 1 mm EDTA, 250 mm imidazole, and protease inhibitor cocktail; MBP vectors: 10 mm HEPES-KOH pH 7.5, 150 mm NaCl, 10 mm maltose, and protease inhibitor cocktail). Proteins were concentrated by the Amicon Ultra filter device (Millipore, UFC501008). All proteins were quantified and aliquoted for storage at −80 °C. Relevant primers are listed in Supplementary Data Set S6.

EMSA

The EMSA assay was conducted with LightShift Chemiluminescent RNA EMSA Kit (Thermo, 89880) according to the protocol described. The 5′-FAM-labeled or 5′-Cy3-labeled probes (10 nm) were incubated with purified proteins (1 μg) at 25 °C for 30 min. Then, samples were loaded to electrophoresis with 0.5 × TBE buffer at 4 °C. Labeled probes were visualized by chemiluminescence apparatus (GE Amersham Imager AI680).

Microscopy imaging

Imaging for rice root tip cells was performed on a Leica SP8 confocal microscope with 40× water objective, Venus was excited at 514 nm and detected at 524 to 557 nm, laser intensity was 3 (CaMV 35S promoter) or 5 (native promoter), the detector was HyD, and gain value was 120%. Imaging for N. benthamiana leaf epidermal cells and protoplast cells was performed on a Leica SP8 confocal microscope with 40× water objective, CFP was excited at 448 nm and detected at 458 to 510 nm, Venus was excited at 514 nm and detected at 524 to 557 nm, CFP and Venus were acquired sequentially to avoid emission crosstalk, CFP laser intensity was 2 and Venus laser intensity was 0.5, the detector was PMT, and gain value was 800. Imaging for droplet was performed on a Leica SP8 confocal microscope with 63× oil objective, CFP was excited at 448 nm and detected at 458 to 510 nm, laser intensity was 0.05, the detector was PMT, and gain value was 650.

In vitro phase separation assay

In vitro phase separation assay was performed as described previously (Zhou et al. 2023). Imidazole in the protein solution was removed by Amicon Ultra filter device, and protein concentrations were determined by the BCA protein quantification kit (Yeasen, 20201ES76). Buffers used in the assay were 20 mm Tris-HCl pH 8.0 and 1 mm EDTA, with different concentrations of NaCl, as indicated in the figure legends. Droplets were assembled in the glass bottom cell culture dish (NEST, 801002), and images were acquired under a confocal microscope (Leica, TCS SP8) equipped with 63× oil immersion objectives. Laser intensity was 0.05. CFP was excited at 448 nm and detected at 458 to 510 nm, the detector was PMT, and gain value was 650. For time-lapse microscopy of CFP-TRBF2 droplet fusion, images were acquired every 2 s for 5 min.

FRAP

FRAP of TRBF2-Venus carrying the ProTRBF2:TRBF2-Venus/WT transgene was performed on a Leica STELLARIS 5 confocal microscope using a 40× water objective. Venus was excited at 514 nm and detected at 524 to 557 nm, laser intensity was 1.5, the detector was HyD, and gain value was 100%. TRBF2-Venus condensates in rice lateral root tip cells were bleached using a laser intensity of 3% at 514 nm, and fluorescence recovery was recorded every 1 s for 30 s after bleaching. FRAP of TRBF2-Venus carrying the Pro35S:TRBF2-Venus/WT transgene was performed on a Leica SP8 confocal microscope using a 40× water objective. Venus was excited at 514 nm and detected at 524 to 557 nm, laser intensity was 1.5, the detector was PMT, and gain value was 800. TRBF2-Venus condensates in rice lateral root tip cells were bleached using a laser intensity of 30% at 514 nm, and fluorescence recovery was recorded every 2 s for 40 s after bleaching. FRAP of AtTRB1-CFP in N. benthamiana leaf epidermal cells was performed on a Leica SP8 confocal microscope using a 40× water objective. CFP was excited at 448 nm and detected at 458 to 510 nm, laser intensity was 1, the detector was PMT, and gain value was 800. AtTRB1-CFP condensates were bleached using a laser intensity of 30% at 448 nm, and fluorescence recovery was recorded every 1 s for 30 s after bleaching.

In vitro FRAP was conducted with samples in the glass bottom cell culture dish using a 63× oil objective of a Leica TCS SP8 confocal microscope. CFP was excited at 448 nm and detected at 458 to 510 nm, laser intensity was 0.05, the detector was PMT, and gain value was 650. A particular region of the CFP-TRBF2 droplet was bleached using a laser intensity of 40% at 448 nm. Fluorescence recovery was recorded every 3 s for 400 s after bleaching.

RNA-seq and data analysis

Two-week-old plants of WT and trbf2-1 without roots were harvested for total RNA extraction. The cDNA libraries were constructed and sequenced using the Illumina Hiseq-PE150 platform by Novogene (China, Beijing). Two independent replicates were analyzed. Raw-sequencing reads were filtered using Trimmomatic (Bolger et al. 2014), and then clean reads were mapped to the O. sativa japonica Group IRGSP-1.0 genome (Ensembl plants) by STAR with default parameters (Dobin et al. 2013). Gene expression was quantified and normalized by TPM (transcripts per kilobase of exon model per million mapped reads). The DEGs were defined using the edgeR package (Robinson et al. 2010) based on the combined thresholds set as the multiple-test corrected P-value (q-value) of 0.05 and cutoff of absolute log₂(fold change) (|log₂FC|) ≥ 1. The R package of VennDiagram (Chen and Boutros 2011) was used to visualize the overlaps between different groups. Visualization of peak distribution along genomic regions of interested genes was performed with Integrative Genomics Viewer (Robinson et al. 2011).

RNA extraction and RT-qPCR assay

Total RNAs were extracted from the shoot of 2-wk-old plants using the hot phenol method (Yang et al. 2017; Xuan et al. 2024) and treated with RNase-free DNase I (Roche, 04716728001) to remove genomic DNA contamination. The full-length cDNA was synthesized by the HiScript II 1st strand cDNA synthesis kit (Vazyme, R211) with oligo dT. Subsequent RT-qPCR processing steps were performed according to the manufacturer's instructions (Vazyme, R311). The OsActin1 gene (Os03g0718100) was used as an internal reference for data normalization. All primers are listed in Supplementary Data Set S6.

ChIP and qPCR assay

The ChIP assay was performed as described previously with modifications (Yang et al. 2017; Zhang, Li, et al. 2022). Briefly, 2-g 2-wk-old plants without roots were ground into fine powder in liquid nitrogen and crosslinked in nuclei isolation buffer I (0.4 m sucrose, 10 mm Tris-HCl pH 8.0, 10 mm MgCl₂, 5 mm β-mercaptoethanol, and 1 mm PMSF) with 1% (v/v) formaldehyde for 10 min at 4 °C. The homogenate was filtered through 2 layers of Miracloth and pelleted by centrifugation. The nucleus was extracted using NIB buffer (20 mm Tris-HCl pH 7.5, 25% [v/v] glycerol, 0.25 m sucrose, 2.5 mm MgCl₂, 10 mm NaCl, 0.5% [v/v] Triton X-100, 0.5 mm spermidine, 0.15 mm spermine, 10 mm β-mercaptoethanol, 1 mm PMSF, and protease inhibitor cocktail) and then resuspended with nuclei lysis buffer (50 mm Tris-HCl pH 8.0, 10 mm EDTA, 1% [w/v] SDS, 1 mm PMSF, and protease inhibitor cocktail) and kept on ice for 30 min. The DNA was sonicated into fragments below 500 bp using Covaris S220. Immunoprecipitations were performed with an anti-H3K27me3 (ABclonal, A2363, Rabbit, 1:500 dilution) and an anti-Flag (Sigma, F3165, Mouse, 1:500 dilution). The bound DNA fragments were then purified after reverse crosslink and amplified by real-time qPCR. To quantify H3K27me3 levels at different chromatin regions, the level at LSI2 (Os03g0107300) was used as the internal control for each sample. Relevant primers are listed in Supplementary Data Set S6.

ChIP-seq and data analysis

The experiment was performed as ChIP-qPCR. Immunoprecipitations were performed with an anti-HA (Cell Signaling, C29F4, Rabbit, 1:500 dilution). The libraries were constructed using the Scale ssDNA-seq Lib Prep Kit for Illumina V2 (ABclonal, RK20228) and sequenced by the Hiseq-PE150 on an Illumina NovaSeq platform at Novogene (China, Beijing). FastQC and MultiQC (Ewels et al. 2016) were used to evaluate the quality of clean reads obtained by sequencing. Clean reads were mapped to the O. sativa japonica Group IRGSP-1.0 genome (Ensembl plants) by Bowtie2 (Trapnell et al. 2012) with default mismatch parameters. The mapping results (in a SAM file) were converted to BAM format using Samtools (Li and Durbin 2009), and only uniquely mapped reads were used for further analysis. Apparent PCR duplicates were removed using Picard tools. MACS2 (Zhang et al. 2008) was employed to call peaks using parameters “gsize = 374931027, f = BAMPE, q = 0.05”. Read counts were CPM (reads of exon model per million mapped reads) normalized using the deepTools (Ramírez et al. 2016) utility bamCoverage with a bin size of 50 bp. Credible peaks were screened using an irreproducible discovery rate (Krismer et al. 2020) and annotated using ChIPseeker (Yu et al. 2015). PlotProfile and plotHeatmap were drawn by deepTools. The R package of VennDiagram (Chen and Boutros 2011) was used to visualize the overlaps between different groups.

Cut&tag and data analysis

Cut&tag assay was performed as described previously with modifications (Kaya-Okur et al. 2020). Briefly, 0.2-g 2-wk-old plants of WT and trbf2-1 without roots were ground into a fine powder and lysed in nucleus extraction buffer (10 mm MES-KOH pH 5.4, 10 mm NaCl, 10 mm KCl, 2.5 mm EDTA, 250 mm sucrose, 0.1 mm spermine, 0.5 mm spermidine, 0.5% [v/v] Triton X-100, 1 mm DTT, and protease inhibitor cocktail). The lysate was filtered through 2 layers of Miracloth and pelleted by centrifugation. Then, nuclei were purified as previously described (Folta and Kaufman 2006). Nuclei were resuspended by 100 μL of wash buffer (20 mm HEPES pH 7.5, 150 mm NaCl, 0.5 mm spermidine, and protease inhibitor cocktail) and incubated with washed 10-μL concanavalin A-coated magnetic beads (Vazyme, TD901) at room temperature for 15 min. Bead-bound nuclei were subjected to library preparation according to the manual of Hyperactive In-Situ ChIP Library Prep Kit for Illumina (pG-Tn5) (Vazyme, TD901). One microgram of rabbit polyclonal antibody anti-H3K27me3 (Millipore, 07-449) or rabbit control IgG (ABclonal, AC005) was used. Two independent replicates were performed.

For library amplification, 24 μL of purified DNA was mixed with 5 μL of ddH₂O, 10 μL of 5× TruePrep Amplify Enzyme buffer, 1 μL of TruePrep Amplify Enzyme (Vazyme, TD901), and 5 μL of P5 and P7 index primers from TruePrep Index Kit V2 for Illumina (Vazyme, TD202). Then, the samples were placed in a PCR machine using the program: 72 °C for 3 min; 98 °C for 30 s; 15 cycles of 98 °C for 15 s, 60 °C for 30 s, and 72 °C for 30 s; 72 °C for 5 min and hold at 4 °C. To purify the PCR products, 1.1× volumes of AMPure XP beads (Beckman, A63880) were added and incubated at room temperature for 15 min. Libraries were washed twice with 80% ethanol and eluted with 22 μL of ddH₂O.

The libraries were sequenced using the Hiseq-PE150 on an Illumina NovaSeq platform at Novogene. Adapters were removed from raw sequencing reads by NGmerge (Gaspar 2018). The reads larger than 130 bp were selected by in-house Python script and further filtered using Trimmomatic. Clean reads were mapped to the O. sativa japonica Group IRGSP-1.0 genome (Ensembl plants) by Bowtie2 (Trapnell et al. 2012) with default mismatch parameters. The mapping results (in a SAM file) were converted to BAM format using Samtools (Li and Durbin 2009), and only uniquely mapped reads were used for further analysis. Apparent PCR duplicates were removed using Picard tools. MACS2 (Zhang et al. 2008) was employed to call peaks using parameters “size = 374931027, bw = 200, q = 0.05”. Read counts were CPM (reads of exon model per million mapped reads) normalized using the deepTools⁵¹ utility bamCoverage with a bin size of 1 bp and extended reads of 200 bp. Credible peaks were screened using an irreproducible discovery rate (Krismer et al. 2020) and annotated using ChIPseeker (Yu et al. 2015). If multiple genes could be assigned to a peak, the closest TSS was selected. DiffBind (Brown and Stark 2012) was used to identify differential enrichment peaks (false discovery rate [FDR] ≤ 0.05) between 2 samples. PlotProfile and plotHeatmap were drawn by deepTools.

Y2H assay

Coding sequences of TRBF2, CLF, and all the truncations were amplified and cloned into the BamH I and Sal I site of the Y2H vectors pGADT7 or pGBKT7. Bait and prey vectors were cotransformed into yeast strain Y2H Gold. The experiments were done according to the Yeast Protocols Handbook. Medium lacking Leu (L), Trp (W), and His (H) (SD-LWH) were used for selection. For spotting on the plates, 5-μL aliquots of yeast cell suspensions (OD₆₀₀ = 1, 0.1, and 0.01) were applied. All primers are listed in Supplementary Data Set S6.

Co-IP

Coimmunoprecipitation (Co-IP) experiments were performed in rice protoplasts. The protoplasts were transfected with 100 μg of each target plasmid and then incubated overnight at 28 °C in the dark. The cultured cells were harvested and lysed by IP buffer (50 mm Tris-HCl pH 7.5, 150 mm NaCl, 1 mm EDTA, 1% [v/v] Triton X-100, 0.1% [w/v] SDS, 5 mm DTT, 1 mm PMSF, and protease inhibitor cocktail). The supernatants were incubated with anti-GFP beads (ChromoTek) at 4 °C for 3 h and then washed 6 times with IP buffer. The protein complex was eluted from the beads by 2× loading buffer (100 mm Tris-HCl pH 6.8, 4% [w/v] SDS, 20% [v/v] glycerol, 0.004% [w/v] bromophenol blue, and 2% [v/v] β-mercaptoethanol). Immunoprecipitates were separated by SDS-PAGE and transferred to a nitrocellulose membrane (GE Healthcare, 10600001). Proteins were detected by immunoblot using the antibodies of anti-HA (ABclonal, AE008, Mouse, 1:2,000 dilution) and anti-GFP (ABclonal, AE078, Rabbit, 1:2,000 dilution).

Sequence alignment and phylogenetic analyses

Homologs of TRBF2 were identified by performing BLAST searches of the protein sequences in NCBI. The protein sequences were aligned with ClustalW (https://www.genome.jp/tools-bin/clustalw) using the default settings, and the Clustal output files were submitted to ESPript 3.0 (https://espript.ibcp.fr/ESPript/cgi-bin/ESPript.cgi) for visualization.

A maximum-likelihood (ML) phylogenetic tree was generated using MEGA11 (Kumar et al. 2018), with default values (substitution type: amino acid; substitution model: Jones–Taylor–Thornton model; rates and patterns: uniform rates; gaps treatment: use all sites; ML heuristic method: nearest neighbor interchange; initial tree for ML: default; branch swap filter: none; number of threads: 3).

Protein alignment and machine-readable files of the phylogenetic analyses are as follows: plant SMH family proteins, Supplementary Files S1 and S2; TRBF2, AtH1, and HsH1 proteins, Supplementary File S3.

Statistical analysis

Statistical analyses were performed using GraphPad Prism version 8. For a 2-group comparison, significance of differences was analyzed by Student's unpaired or paired 2-tailed t-tests. For multiple comparisons, in case of equal standard deviations, 1-way ANOVA (nonparametric) tests were performed with Tukey's multiple comparisons. Statistical tests and the number of replicates are indicated in the figure legends; all statistical data are provided in Supplementary Data Set S7.

Accession numbers

Sequence data from this article can be found at Ensembl Plants under the following accession numbers: TRBF2 (Os12g0613300), TRBF1 (Os01g0589300), TRBF3 (Os01g0708000), MADS27 (Os02g0579600), SGD2 (Os01g0643600), CMT3a (Os10g0104900), CLF (Os06g0275500), RTBP1 (Os02g0817800), and TRFL1 (Os02g0776700).

The RNA-seq and Cut&tag data sets supporting the conclusions of this paper have been deposited in the Sequence Read Archive (SRA) under accession PRJNA811378. The ChIP-seq data sets supporting the conclusions of this paper have been deposited in the SRA under accession PRJNA948887. All data supporting the findings of this study are available from the corresponding author upon reasonable request.

Author contributions

H.Y. and H.X. designed the study; H.X. performed and supervised all experimental works; Y. Li and Y. Liu carried out the bioinformatic analysis; J.Z., J.C., and N.S. helped to generate plant materials; L.P. and S.L. helped for the phenotype measurements; G.X. and Y.Z. helped with the phase separation assay; and H.Y., H.X., and Y, Liu wrote the manuscript.

Supplementary data

The following materials are available in the online version of this article.

Supplementary Figure S1. Amino acid sequence alignment of plant SMH family proteins.

Supplementary Figure S2. Grain size of TRBF2 mutants.

Supplementary Figure S3. TRBF2 has the telomere repeat-binding activity.

Supplementary Figure S4. RNA-seq analysis of WT and trbf2.

Supplementary Figure S5. TRBF2 ChIP-seq data analysis.

Supplementary Figure S6. TRBF2 forms phase-separated condensates in vivo.

Supplementary Figure S7. TRBF2 forms phase-separated condensates in vitro.

Supplementary Figure S8. Arabidopsis AtTRB1 can display phase-separated characteristics.

Supplementary Figure S9. Characterization of TRBF2 protein.

Supplementary Figure S10. The phase separation detection of TRBF2 and truncations.

Supplementary Figure S11. Protein levels detected by western blot in Fig. 6C.

Supplementary Figure S12. TRBF2ΔH1/H5 and TRBF2-IDR^AtSE ChIP-seq data analysis.

Supplementary Figure S13. Mapping of TRBF2 and CLF interaction domains.

Supplementary Figure S14. The phase-separated TRBF2-IDR^AtSE attracts CLF in vitro and in planta.

Supplementary Table S1. Summary of RNA-seq, ChIP-seq, and Cut&tag data.

Supplementary Data Set S1. List of the significantly misregulated genes in trbf2 compared with WT.

Supplementary Data Set S2. List of the TRBF2-binding genes.

Supplementary Data Set S3. List of the TRBF2ΔH1/H5-binding genes.

Supplementary Data Set S4. List of the TRBF2-IDR^AtSE-binding genes.

Supplementary Data Set S5. List of the H3K27me3 altered regions and genes in trbf2 compared with WT.

Supplementary Data Set S6. List of the primers used in this study.

Supplementary Data Set S7. Summary of statistical analyses.

Supplementary File S1. Machine-readable tree file of plant SMH family proteins.

Supplementary File S2. Multiple sequence alignments of plant SMH family proteins.

Supplementary File S3. Multiple sequence alignments of TRBF2, AtH1, and HsH1 proteins.

Funding

This project was supported by the Fundamental Research Funds for the Central Universities (2042022rc0007), the National Natural Science Foundation of China (32370619 and 31871301), the Natural Science Foundation of Hubei Province (2024AFA010), Hubei Hongshan Laboratory, and a start-up fund from Wuhan University.

Dive Curated Terms

The following phenotypic, genotypic, and functional terms are of significance to the work described in this paper:

• ATS Gramene: AT5G42630

• ATS Araport: AT5G42630

• EMB1579 Gramene: AT2G03150

• EMB1579 Araport: AT2G03150

• AAT Gramene: AT2G22250

• AAT Araport: AT2G22250