Glycine max Jumonji C domain-containing proteins JMJ19/20 exhibit endopeptidase activity and interact with LUXs to mediate flowering time
Yapeng Han, Qian Jia, Mengshi Liu, Yuan Fang, Tianqi Shen, Shiru Tan, Qianya Gao, Jing Liu, Chenjiang You, Yingxiang Wang
TL;DR
This study explores how two soybean proteins, GmJMJ19 and GmJMJ20, influence flowering time by acting as endopeptidases and interacting with another protein called LUX.
Contribution
The novel finding is that GmJMJ19/20 function as endopeptidases rather than histone demethylases and influence flowering time through a non-canonical pathway.
Findings
GmJMJ19/20 cleave histone H3 peptides at unmethylated lysine 27 residues without requiring Fe²⁺ or α-ketoglutarate.
Knockout of GmJMJ19/20 significantly delays flowering time in soybean.
GmJMJ19/20 interact with GmLUX2 and enhance GmFULc expression independently of the canonical evening complex.
Abstract
Histone demethylases serve essential functions in plant growth and development across various species. However, their roles in Glycine max remain largely unexplored. This study identified GmJMJ19 and GmJMJ20, encoding JmjC domain-containing proteins. They exhibit circadian rhythmic expression patterns and interact with LUX ARRHYTHMO 2 (GmLUX2) both in vitro and in vivo. Although GmJMJ19/20 bind to histones, they lack conventional histone demethylase activity. Rather, GmJMJ19/20 function as endopeptidases that specifically cleave histone H3 peptides at unmethylated lysine 27 residues, in a manner independent of Fe²⁺ and α-ketoglutarate, the cofactors required for typical JmjC enzymes. Structural modeling supports the occlusion of the catalytic pocket that prevents access to methylated substrates. The simultaneous knockout of GmJMJ19/20 significantly delays flowering time. Comparative…
Genes, proteins, chemicals, diseases, species, mutations and cell lines named across the full text — each resolved to its canonical identifier and authoritative record.
Click any figure to enlarge with its caption.
Figure 1
Figure 2
Figure 3
Figure 4
Figure 5
Figure 6
Figure 7
Figure 8- —National Science and Technology Major Project10.13039/501100018537
- —National Natural Science Foundation of China10.13039/501100001809
- —Guangdong Pearl River Talent Program
- —South China Agricultural University10.13039/501100012601
- —Basic and Applied Basic Research Foundation
- —South China Agricultural University10.13039/501100012601
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsPlant Molecular Biology Research · Soybean genetics and cultivation · Plant Gene Expression Analysis
Introduction
Epigenetic regulation through post-translational modifications (PTMs) of histones plays a fundamental role in modulating chromatin structure and gene expression [1]. The nucleosome, the basic unit of chromatin, consists of ~146 bp of DNA wrapped around a histone octamer containing two copies each of H2A, H2B, H3, and H4 [2]. The N-terminal tails of histones are subjected to diverse PTMs, including methylation, acetylation, phosphorylation, ubiquitination, and SUMOylation [3], which are dynamically regulated by specific histone-modifying enzymes and influence transcriptional activity and chromatin accessibility. Among histone demethylases, Jumonji C (JmjC) domain-containing proteins (JMJs) represent the largest family, capable of removing tri-, di-, and monomethyl groups from lysine residues [4, 5]. In the model plant Arabidopsis thaliana, there are 21 JMJs proteins, while only 20 are reported in rice and tomato [6–8]. Based on domain architecture and sequence similarity, plant JMJs are classified into five major groups: KDM3, KDM4, KDM5, JMJD6, and the JmjC domain-only group [4, 6]. Notably, members of the JmjC domain-only group have diverged evolutionarily in higher eukaryotes and may function independently of histone demethylation [5].
Histone tails, whether intrinsically disordered or modified, substantially influence genome stability, accessibility, and dynamics. In recent years, histone clipping—proteolytic cleavage of histones tails—has emerged as a novel layer of chromatin regulation. JMJD5 and JMJD7, members of the JmjC domain-only group, have been identified as histone-specific proteases [9, 10]. Previous studies indicated that HsJMJD5 lacks demethylase activity [11–15], but instead exhibits hydroxylase and endopeptidase activity targeting the histone H3 N-terminal tail [9, 10, 16, 17]. Although histone clipping has been documented from yeast to mammals, it remains largely uncharacterized in plants.
Soybean serves as a globally significant crop, providing a primary source of plant protein and oil. Domesticated ~5000 years ago from its wild progenitor Glycine soja (Sieb. and Zucc.) in the Huang-Huai Valley of central China, soybean is a facultative SD plant exhibiting high sensitivity to photoperiod [18]. Photoperiodic control of flowering is a key determinant of regional adaptation and yield potential in soybean [19]. Recent research has identified numerous flowering time regulators that establish a flower time control network, wherein E1 functions as the major floral repressor of soybean photoperiodic flowering pathway [20]. The evening complex (EC) and time of flowering 11 and time of flowering 12 (Tof11/Tof12) constitute two critical pathways that perceive signals from photoreceptors E3 and E4 to modulate the expression of E1 [21–24]. E1 encodes a B3 domain transcription factor, and has two paralogs, E1 like a (E1La) and E1 like b (E1Lb). All three proteins bind to the promoters of two FLOWERING LOCUS T (GmFT2a and GmFT5a) to repress their expression, thus delaying flowering [25, 26]. Additionally, E1 suppresses the transcription of FRUITFULLc/Time of Flowering 5 (GmFULc/Tof5), also known as GmFUL2a and MADS-box genes downregulated by E1 05 (GmMDE05), while recent study demonstrated that GmFULc can indirectly repress E1 and E1Lb [27, 28].
This study identified two paralogous JmjC domain-containing genes GmJMJ19/20 in soybean that exhibit circadian rhythmic expression patterns, similar to the EC component GmLUX2. Genetic analysis reveals that the double mutant gmjmj19 gmjmj20 (hereafter designated as gmjmj1920) exhibits moderately delayed flowering compared to wild-type control under SD conditions. Biochemical analyses demonstrate that GmJMJ19/20 possess histone H3 tail clipping activity, specifically at unmethylated lysine 27 (H3K27), while lacking canonical demethylase activity. This endopeptidase activity operates independently of Fe²⁺ and α-ketoglutarate (α-KG), which are essential cofactors for typical JmjC demethylases. Structural modeling using AlphaFold3 indicates that GmJMJ19/20 contain conserved substrate-occluding residues within the catalytic pocket, similar to HsJMJD5, suggesting restricted access to methylated lysines and supporting a noncanonical enzymatic mechanism. Additionally, GmJMJ19/GmJMJ20 and GmLUX2 cooperatively regulate multiple flowering genes including GmFULc, promoting its expression likely through histone H3 tail clipping in promoter regions. Analysis of natural variation suggests that GmJMJ19 underwent selection during domestication, contributing to adaptation in low-latitude environments. These findings reveal a novel role for GmJMJ19/20 as potential histone-clipping enzymes in soybean flowering time regulation, offering new targets for molecular breeding to expand soybean cultivation across diverse geographic regions.
Materials and methods
Plant material and growth conditions
Unless otherwise mentioned in the main text, soybean cultivars in the Williams 82 background were cultivated in a plant growth chamber under SD (12 h light/12 h dark), LD (16 h light/8 h dark), or constant light (24 h light) at 25°C. Nicotiana benthamiana plants were cultivated at 22°C under 16 h light/8 h dark conditions.
Phylogenetic tree
The JmjC domain sequences from selected JmjC domain containing proteins, OsJMJ711 (Q75LR4), OsJMJ712 (A3C049), OsJMJ714 (A3C030), OsJMJ709 (Q5JKD1), OsJMJ713 (Q5ZC07), OsJMJ717 (LOC_Os09g31380), HsPKDM11 (NP_075383.2), HsJMJD5 (NP_079049.2), HsPKDM12B (NP_060372.2), HsPKDM12C (NP_078886.2), and HsJMJD7 (NP_001108104.1) from NCBI, and AtJMJ30 (AT3G20810), AtJMJ31 (AT5G19840), AtJMJ20 (AT5G63080), AtJMJ32 (AT3G45880), GmJMJ21 (Glyma.02g144300), GmJMJ22 (Glyma.10g029800), GmJMJ23 (Glyma.11g089600), and GmJMJ24 (Glyma.04g185900) were obtained from Phytozome v13 (https://phytozome-next.jgi.doe.gov/). Amino acid sequences were aligned using ClustalW, and the phylogenetic tree was constructed and visualized in MEGA 11 [29] utilizing the JTT model, pairwise deletion, and neighbor-joining method for 1000 bootstrap replications.
Identification of SNPs in GmJMJ19 and GmJMJ20 and geographical distribution
Whole genome resequencing data of 2898 accessions (PRJCA002030) was obtained from the national genomics data center (https://ngdc.cncb.ac.cn/). After filtering out SNPs with missing rates exceeding 10% and minor allele frequency below 5% in VCF format file, the remaining Single Nucleotide Polymorphisms (SNPs) were utilized for genetic diversity analysis. VCFtools [30] was employed to calculate the pairwise genomic differentiation values for wild, landrace, and improved cultivar soybean populations per 10 kb sliding window, including the fixation index (Fst) of population differentiation, nucleotide diversity (π), and Tajima’s D, which indicates whether a population is expanding or shrinking. BCFtools [31] was utilized to filter SNPs in the gene body of GmJMJ19 gene for haplotype identification. Accessions collected in PRC were classified accordingly and mapped to their geographical origins. The map of PRC was certified and obtained from the Ministry of Natural Resources of PRC (https://www.bzdt.ch.mnr.gov.cn).
RNA extraction, RT-qPCR, and RNA sequencing
Total RNA from soybean tissues was extracted using RN38-EASYspin Plus (Aidlab, RN38) according to the manufacturer’s protocol. PrimeScript™ RT reagent Kit with gDNA Eraser (Perfect Real Time) (TaKaRa, #RR047A) was utilized to synthesize first-strand complementary DNAs (cDNAs) from the total RNA. Reverse Transcription Quantitative Polymerase Chain Reaction (RT-qPCR) assays were performed using qPCR SYBR Green Master Mix (YEASEN, 11198ES08) on a CFX96 Connect Real-Time PCR Detection System (Bio-Rad). Each experiment was conducted at least three times with similar results. The primers used are listed in Supplementary Table S1.
Library construction and RNA sequencing (RNA-seq) were performed by Shanghai Majorbio Bio-pharm Technology Company (Shanghai, China) using total RNA from 14 DAE soybean leaves. pRNASeqTools (https://github.com/grubbybio/pRNASeqTools) was employed to analyze the sequencing results. Raw reads were processed using cutadapt v4.2 to trim 3′ adapters and low-quality tails. The remaining reads were aligned to the soybean genome Wm82.a2.v1 [32] using STAR v2.7.10b with default parameters. featureCounts v2.0.3 [33] was employed to count mapped reads of each gene, and fragments per kilobase of exon model per million mapped fragments was used to estimate gene expression levels. Differentially expressed genes (DEGs) were identified using DESeq2, with an absolute value of log_2_(fold change) ≥1.5 and *P-*value <.05 as threshold criteria.
Yeast two-hybrid assay
The full-length, N-terminus, and C-terminus sequences of GmLUX2 CDS (W82) were amplified and inserted into pGBKT7 (BD vector). Similarly, the full-length, N-terminus, and C-terminus sequences of GmJMJ19 and GmJMJ20 CDS were amplified and inserted into pGADT7 (AD vector). The pGBKT7 vectors containing CDS sequences of other EC members were previously described [23]. All AD vectors were transformed into Yeast Strain Y187, while all BD vectors were transformed into Yeast Strain Y2HGold. The transformants were mated on YPDA medium, and interactions were evaluated on double dropout supplement [DDO, prepared using dropout supplement (Takara 630428, Clontech) with adenine and histidine], and quadruple dropout supplement (QDO, dropout supplement, Takara 630428, Clontech) medium.
Pull-down and electrophoretic mobility shift assay
For the pull-down assay, GmJMJ19 and GmJMJ20 CDS were inserted into pET-28a(+)-SUMO (His tag). The GmLUX2 CDS was inserted into pGEX6T-1 (GST tag), as previously reported [23]. Recombinant proteins were expressed in Escherichia coli Rosetta2 (DE3) cells. GST Bind Resin (Merck, Germany) and Ni-NTA (Merck, Germany) were utilized for protein purification. GmJMJ19-His or GmJMJ20-His proteins were incubated with 40 μl Ni-NTA in 1 ml pull down buffer at 4°C for 1 h. After washing, the beads were split into two equal portions and incubated with either GST or GmLUX2-GST for 2 h. The washed beads were then boiled in 100 μl 1 × sodium dodecyl sulphate (SDS) loading buffer, and supernatants underwent western blot analysis using anti-His (M20001, Abmart, Shanghai, China) or anti-GST (M20007, Abmart, Shanghai, China) antibodies.
GmLUX2- His-SUMO proteins were employed for electrophoretic mobility shift assays (EMSAs), with His-SUMO serving as control. The probes (5′ FAM-labeled WT probes, nonlabeled cold probes, and mutant probes) for EMSA were synthesized by Genewiz, Suzhou (Supplementary Table S1). The purified proteins were incubated with 0.5 μM labeled probes in a 20-μl reaction mixture containing 10 mM Tris−HCl, pH 7.5, 0.05% Nonidet P-40, 10 mM MgCl_2_, 5% (v/v) glycerol, 0.5 μg/μl bovine serum albumin, and 1 mM Dithiothreitol (DTT). The reactions were incubated at 25°C for 40 min, followed by separation on 6% native polyacrylamide gels in 0.5× Tris-Borate-EDTA (TBE) buffer. The gel imaging was imaged using Chemiluminescence instrument (Tanon-5200).
Split-LUC assay and bimolecular fluorescence complementation assay
For the split-LUC assay, the CDS of GmJMJ19, GmJMJ20, and GmLUX2 were inserted into 35S:nLUC and 35S:cLUC vectors [34]. For the bimolecular fluorescence complementation assay (BiFC) assay, the coding sequences of GmJMJ19, GmJMJ20, and GmLUX2 were inserted into pXY106 and pXY104 vectors [35]. The Agrobacterium strain GV3101 carrying corresponding vectors in equal OD values were mixed and injected into young leaves of N. benthamiana. After 36 h, the leaves for split-LUC assay and BiFC assay were observed under LB985 NightShade with indiGo software (Berthhold Tech) or LSM-710 confocal microscope (Zeiss, Germany), respectively.
Co-immunoprecipitation assay
Full-length coding sequences of GmLUX2, GmJMJ19, and GmJMJ20 were amplified from W82 cDNA and introduced into the 35S:MYC and 35S:GFP vector, respectively. Different combinations of GV3101 carrying indicated constructions were cotransformed into N. benthamiana leaves. After 36 h, total protein was extracted from the corresponding leaves with IP buffer [50 mM Tris−HCl, pH 8.0, 100 mM NaCl, 5 mM MgCl_2_, 0.1% Nonidet P-40, 10% glycerol, 2 mM DTT, and 1× ethylenediaminetetraacetic acid (EDTA)-free protease inhibitor cocktail]. The samples were incubated at 4°C for 30 min, followed by centrifugation twice at 13 000 × g for 5 min at 4°C. The supernatants were gently mixed with 20 μl GFP trap beads (ChromoTek, gta-20) at 4°C for 2 h. Following incubation, beads were collected by gentle centrifugation and washed five times with 1 ml IP buffer. The beads were then diluted with 100 μl 1× SDS buffer and heated at 95°C for 5 min. Supernatants were analyzed by sodium dodecyl sulfate−polyacrylamide gel electrophoresis and western blotting, using corresponding antibodies.
Chromatin preparation and detection of histone H3 clipping by western blotting
Chromatin was isolated from the leaves of 14-day-old plants (W82, gmjmj1920-1 and gmjmj1920-2) following established protocols with modifications, specifically, the omission of protease inhibitors throughout the process [36, 37]. Leaf tissues were frozen in liquid nitrogen, ground into a fine powder, and homogenized in nuclear extraction buffer (20 mM Tris−HCl, pH 7.5, 20 mM KCl, 2 mM EDTA, 2.5 mM MgCl₂, 250 mM sucrose, 5 mM DTT, 25% glycerol). The homogenate was filtered through Miracloth and centrifuged at 1500 × g for 5 min. The pellet was washed five times with NRBT buffer (20 mM Tris−HCl, pH 7.5, 2.5 mM MgCl₂, 0.2% Triton X-100, 25% glycerol). The washed nuclear pellet was then resuspended in NRB2 buffer (20 mM Tris−HCl, pH 7.5, 250 mM sucrose, 10 mM MgCl₂, 0.5% Triton X-100, 5 mM β-mercaptoethanol) and carefully overlaid on the equal volume of NRB3 cushion buffer (20 mM Tris−HCl, pH 7.5, 1.7 M sucrose, 10 mM MgCl₂, 0.5% Triton X-100, 5 mM β-mercaptoethanol). Following centrifugation at 12 000 × g for 45 min at 4°C, the purified nuclei were collected from the pellet. For chromatin extraction, the nuclear pellets were resuspended in 500 μl of glycerol buffer (20 mM Tris−HCl, pH 7.5, 75 mM NaCl, 0.5 mM EDTA, 1 mM DTT, 50% glycerol, 10 mM β-mercaptoethanol) and gently overlaid onto 500 μl urea buffer (10 mM HEPES, pH 7.5, 1 mM DTT, 7.5 mM MgCl_2_, 0.2 mM EDTA, 0.3 M NaCl, 1 M urea, 1% Nonidet P-40). The tubes were vortexed gently twice for 2 s, incubated on ice for 5 min, and centrifuged at 12 000 × g for 5 min at 4°C. The final chromatin pellets were obtained for subsequent analysis. The chromatin samples were solubilized in 1× SDS loading buffer by boiling at 95°C for 5 min. Western blotting was performed using an antibody against the C-terminal of histone H3 (Abcam, ab1791). Histone H4 was detected by anti-H4 (Abcam, ab10158) in parallel as a loading control.
Generation of transgenic plants
For CRISPR/Cas9, Single-guide RNAs (sgRNAs) were designed using the web tool CRISPR-P 2.0 [38]. Pairs of DNA oligonucleotides for the sgRNAs were synthesized by Genewiz and annealed to generate dimers, which were subsequently integrated into the pGES201 vector as previously reported [39]. The soybean hairy root transformations and stable transformations were conducted by Wuhan Edgene Bio-tech co., LTD.
The 2000-bp promoter region plus full-length CDS sequence of GmJMJ20 were amplified from W82 genomic DNA and cDNA, respectively, and subsequently introduced into the pTF101 vector, which uses BASTA as the selection marker. The plasmid was introduced into Agrobacterium tumefaciens strain EHA101 and subsequently transformed into soybean W82 following the cotyledon-node method [40].
Protein structure prediction and visualization
The three-dimensional structures of GmJMJ19, GmJMJ20, and AtJMJ30 were predicted using AlphaFold3 (https://alphafoldserver.com). The structural superimpositions and visualization were performed using PyMoL (version 3.1.3, Schrödinger, LLC).
Chromatin immunoprecipitation
Leaf tissue was collected from 14-day-old plants at ZT12 under SD conditions from W82 and GmJMJ20pro:GmJMJ20-3 × flag plants. Approximately 2 g samples were fixed using 1% formaldehyde for 25 min in vacuo, followed by 7 min quenching with 125 mM glycine. Nuclei were isolated from these leaves and sonicated with a Bioruptor Plus sonication device UCD300 (Diagenode) to obtain DNA fragments of ~500 bp. Anti-AtLUX (Abiocode, USA) and FLAG M2 Magnetic Beads (Sigma, M8823) were used for immunoprecipitation. DNA fragments were recovered through phenol–chloroform extraction. Immunoprecipitated DNA was used for chromatin immunoprecipitation (ChIP)-seq DNA library preparation or qPCR analysis.
Two biological replicates were prepared and sequenced for each ChIP-seq experiment. End repair, adaptor ligation, and amplification were performed using the VAHTS^®^ Universal DNA Library Prep Kit (ND607) according to the manufacturer’s protocol. The Novaseq 6000 was used for high-throughput sequencing of the ChIP-seq libraries at Novogene (Tianjin). The raw sequence data were processed using pRNASeqTools. Initially, raw reads were trimmed by cutadapt v4.2 [41]. Bowtie2 was then employed to map the reads to soybean Wm82.a2.v1 reference genome. Peaks were called by Genrich v0.6.1 with defined threshold value (fold change ≥4 and *P-*value <.01), and credible peaks were annotated using the R package ChIPseeker [42]. The data were subsequently imported into the Integrated Genome Viewer (IGV) [43] for visualization.
For ChIP-qPCR assay, the precipitated DNA was analyzed by quantitative real-time PCR in triplicate each biological replicates. The enrichment of the EF1b genomic fragment served as a negative control. The percentage of each amplicon was determined as the ratio between immunoprecipitated DNA and input DNA. All primers are listed in Supplementary Table S1.
CUT&tag assay
The soybean leaves were sampled from 14-day-old seedling at ZT12 under SD. The nuclei were extracted using Plant Nuclei Isolation/Extraction Kit (Sigma, CELLYTPN1). Then the CUT&Tag assay was conducted following the manual of Hyperactive Universal CUT&Tag Assay Kit for Illumina (Vazyme, TD903) with antibodies against histone H3 and three major histone methylation marks: H3K4me3 (Millipore, 07-473), H3K27me3 (Millipore, 07-449), H3K36me3 (Abcam, ab9050), and H3 (Abcam, ab1791). Two biological replicates were prepared and sequenced for each CUT&Tag experiment. The enriched DNA was purified by VAHTS DNA Clean Beads (Vazyme, N411), and libraries were constructed using library prep kit (Vazyme, TD202) according to the manufacturer’s protocol. The Novaseq 6000 was used for high-throughput sequencing of the CUT&Tag libraries at Genergy Biological Technology Limited Co. (Shanghai, China). The raw data processing and further analysis were the same as ChIP-seq analysis. The meta-plots were draw by deepTools (version 2.5), and *P-*values were calculated by ANOVA comparing fitted models [44].
Demethylation activity assay in vitro
The in vitro demethylation assay was conducted according to previously established protocols [45, 46]. GmJMJ19 and GmJMJ20 were expressed as recombinant proteins using the pET-28a (+)-SUMO vector, while AtJMJ30 was expressed using the pMAL-c2X vector containing an N-terminal His-MBP tag. Mutants were generated using the QuikChange site-directed mutagenesis kit (Stratagene) with wild-type plasmids as templates. All primers are listed in Supplementary Table S1. Each recombinant vector was transformed into E. coli Rosetta2 (DE3) cells for target protein overexpression. All recombinant proteins underwent initial purification by Ni-NTA affinity chromatography (Qiagen), followed by ion exchange purification using a Q HP column (GE Healthcare). SUMO tags were removed by Ulp1 protease, and His-MBP tags were removed by TEV protease. The cleaved proteins underwent further purification by HisTrap affinity chromatography (GE Healthcare) to obtain high-purity proteins for subsequent demethylation assays. A quantity of 50 µM synthesized peptides with different H3 modifications (Scilight peptide, Beijing, China) was incubated with 10 µM proteins in 40 μl reaction buffer (20 mM Tris−HCl, pH 7.5, 150 mM NaCl, 50 µM Fe[NH_4_]2[SO_4_]2, 100 µM α-KG, and 50 µM ascorbate) for 3 h at 37°C and 25°C, respectively. The reaction mixture was further subjected to a MALDI-TOF/TOF mass spectrometer (Bruker ultrafleXtreme) at the Analysis and Testing Center, Sun Yat-sen University.
Statistical analysis
In addition to sequencing data, other data underwent statistical analysis using GraphPad Prism 8.0 for calculating the mean and standard deviation. Two-sided Student’s t-test was performed using Microsoft Excel 2021 (Microsoft, USA).
Gene accession
Sequence data for soybean genes described in this paper can be found in SoyBase (https://www.soybase.org) under the following accession numbers: GmJMJ19 (Glyma.12G055000), GmJMJ20 (Glyma.11G130600), E1 (Glyma.06G207800), GmELF3a/J (Glyma.04G050200), GmELF3b-1 (Glyma.14G091900), GmELF3b-2 (Glyma.17G231600), GmELF4a (Glyma.11G229700), GmELF4b (Glyma.07G037300), GmLUX1 (Glyma.12G060200), GmLUX2 (Glyma.11G136600), GmFT2a (Glyma.16G150700), GmFT5a (Glyma.16G044100), GmFULc (Glyma.05G018800).
Results
GmJMJ19/20 are paralogs and coexpress with GmLUX2
Previous studies have demonstrated that GmLUX2 regulates soybean flowering time [22, 23]. To identify additional genes potentially involved in this process, a coexpression analysis was conducted using published soybean expression datasets through ATTED-II (https://atted.jp/). Among the top 20 genes coexpressed with GmLUX2 (Supplementary Table S2), several known EC components including GmELF3a, GmLUX1, and GmELF4b were identified, validating the analysis reliability. Further investigation focused on the top-ranked gene Glyma.12G055000, annotated as a homolog of Arabidopsis gene AtJMJ30, which encodes a lysine-specific demethylase. Another homolog of AtJMJ30, Glyma.11G130600, ranked 16th. Based on previous gene family classification [47], these two genes shared high amino acid sequence similarity (Supplementary Fig. S1) and were designated as GmJMJ19 (Glyma.12G055000) and GmJMJ20 (Glyma.11G130600), respectively. Phylogenetic analysis placed GmJMJ19/20 within the JmjC domain-only subfamily, closely related to AtJMJ30, OsJMJ712, and HsJMJD5 (Supplementary Fig. S2).
Given the rhythmic expression of GmLUX2 [22, 23, 48] and its coexpression with GmJMJ19/20, these genes were hypothesized to exhibit circadian expression patterns. Leaf samples from plants grown under SD and LD photoperiods were collected every 4 h over a 48-hour period for real-time PCR analyses. Both genes exhibited robust diurnal rhythms with peak expression at ZT12 under both SD and LD conditions (Fig. 1A and B), whereas GmJMJ19 showed substantially lower expression than that of GmJMJ20 under LD conditions. The different patterns may suggest partial sub-functionalization or compensation between the two paralogs. Under constant light, their rhythmic expression persisted (Fig. 1C), indicating circadian clock regulation of transcription levels, similar to GmLUX2 (Supplementary Fig. S3) [22, 23, 49]. Tissue-specific expression analysis revealed that GmJMJ19 and GmJMJ20 are predominantly expressed in leaves compared to other tissues examined (Fig. 1D). This expression pattern is consistent with in vivo function participating in photoperiod-dependent flowering time control. GFP-fused GmJMJ19/20 driven by the CaMV 35S promoter were localized to both nucleus and cytoplasm in N. benthamiana (Fig. 1E). The expression patterns and localizations were both consistent with their homologs AtJMJ30 in Arabidopsis and OsJMJ713 in rice [49–51].
Analyses of the expression GmJMJ19 and GmJMJ20 and their protein subcellular localization. The expression patterns of GmJMJ19 and GmJMJ20 in SD (A), LD (B), and constant light conditions (C). The peaks of GmJMJ19 and GmJMJ20 expression appear at ZT12 under all conditions. Gray and blue background represent night and subjective night, respectively. Error bars stand for standard deviations of three technical replicates. Each experiment was performed three times with similar results. (D) The expression patterns of GmJMJ19 and GmJMJ20 under SD in examined tissues. Both genes exhibit highest expression in leaves. GmACT served as the internal control. Error bars represent standard deviations of three biological replicates. (E) The subcellular localization of GmJMJ19 and GmJMJ20. Fluorescent signals of GmJMJ19-GFP and GmJMJ20-GFP were detected in both nucleus and cytoplasm. Bar = 50 μm.
GmJMJ19/20 interact with GmLUX2 both in vitro and in vivo
Considering that coexpressed genes typically share related biological functions [52], we hypothesized that GmJMJ19/20 might interact genetically and/or physically with GmLUX2. To examine physical interactions, we initially conducted in vitro pull-down assays using His- and GST-tagged fusion proteins. GmJMJ19/20-His effectively pulled down GmLUX2-GST, but not the GST control (Fig. 2A), indicating a direct interaction. To confirm these interactions in vivo, we performed split luciferase complementation (split-LUC) assays in transiently transformed tobacco (N. benthamiana) leaves. Substantial luciferase activity was detected when nLUC-GmJMJ19/20 was coexpressed with GmLUX2-cLUC (Fig. 2B). Bimolecular fluorescence complementation (BiFC) assays subsequently localized their interactions to the nucleus, wherein GmJMJ19/20 also interacted with GmLUX1, a homolog of GmLUX2 (Supplementary Fig. S4A). The in vivo interactions were additionally verified through co-immunoprecipitation (Co-IP) assays in tobacco leaves transiently coexpressing GmJMJ19/20-GFP with GmLUX2-MYC, and western blot analysis demonstrated that GmLUX2-MYC coprecipitated with both GmJMJ19/20-GFP (Fig. 2C). To determine the specific interacting domains, we implemented yeast two-hybrid (Y2H) assays using truncated GmJMJ19/20 and GmLUX2 proteins (Fig. 2D). Since the full-length GmLUX2 and its N-terminus exhibited self-activation in Y2H system (Supplementary Fig. S4B), we established that the C-terminus of GmLUX2 lacking self-activation interacts with the C-terminal regions of both GmJMJ19 and GmJMJ20 (Fig. 2E). No interactions were detected between GmJMJ19/20 and other soybean EC components, indicating interaction specificity (Supplementary Fig. S4C). Furthermore, BiFC assays revealed that GmJMJ19 and GmJMJ20 can interact with each other in tobacco cells (Supplementary Fig. S4D and E), suggesting that GmJMJ19 and GmJMJ20 might form heterodimers in vivo.
GmJMJ19 and GmJMJ20 interact with GmLUX2. (A) GmJMJ19 and GmJMJ20 interact with GmLUX2 in vitro. GmJMJ19-HIS and GmJMJ20-HIS served as baits, while GmLUX2-GST and GST served as preys. GST was utilized as the negative control. (B) GmJMJ19 and GmJMJ20 interact with GmLUX2 in tobacco leaves. GmJMJ19 and GmJMJ20 were fused with nLUC, whereas GmLUX2 was fused with cLUC. The bioluminescence signaling intensity was detected with a CCD camera. (C) Co-IP assays showing both GmJMJ19 and GmJMJ20 interact with GmLUX2 in tobacco leaves. Total proteins were extracted and immunoprecipitated with GFP-Trap beads. Immunoblot detection was conducted using GFP or MYC antibodies as indicated. (D) Schematic diagram illustrates the domain architecture of GmJMJ19, GmJMJ20, and GmLUX2 and depicts the designs of truncated proteins for protein−protein interaction assays. (E) Y2H assay demonstrates that both C-termini of GmJMJ19 (GmJMJ19-C) and GmJMJ20 (GmJMJ20-C) can interact with the C-terminus of GmLUX2 (GmLUX2-C). The full-length and truncated forms of GmJMJ19 and GmJMJ20 were fused with AD, while GmLUX2-C was fused with BD, respectively.
GmJMJ19 and GmJMJ20 exhibit histone H3 endopeptidase activity
Based on their JmjC domain composition and sequence similarity to known demethylases, GmJMJ19/20 were initially hypothesized to function as histone demethylases. To examine this, we first assessed whether GmJMJ19/20 could bind nucleosome. Pull-down assays using calf thymus histones as bait showed that both GmJMJ19 and GmJMJ20 efficiently pulled down histone H3 and H4, but not H1 or H2 (Supplementary Fig. S5). We subsequently evaluated their binding affinity to histone H3 peptides bearing four types of trimethyl-lysine modifications: H3K4me3, H3K9me3, H3K27me3, and H3K36me3. Isothermal titration calorimetry (ITC) results indicated that GmJMJ19 binds all four peptides (Fig. 3A–D and Supplementary Table S3).
GmJMJ19 and GmJMJ20 exhibit histone H3 endopeptidase activity. ITC assays showing the binding affinities of GmJMJ19 to H3K4me3 (A), H3K9me3 (B), H3K27me3 (C), and H3K36me3 (D) peptides. Experiments were performed with three independent protein preparations, and representative binding curves are shown. Dissociation constants (Kd) are reported as mean ± SD (n = 3). (E−H) MALDI-TOF mass spectrometry analysis of enzymatic reactions using the corresponding trimethylated peptides as substrates. Demethylated products were not detected; however, clipping products were observed in reactions involving H3K36me3 with either GmJMJ19 or GmJMJ20. The red arrow indicates the cleavage site, while the red and blue lines represent the products and substrates, respectively. Negative controls (−) included reactions without enzyme, none of which showed detectable cleavage. All enzymatic assays were repeated in triplicate with consistent results.
We then investigated whether GmJMJ19/20 possessed histone demethylase activity. Each protein was incubated with the same set of trimethylated histone H3 peptides, followed by MALDI-TOF mass spectrometry analysis (Fig. 3E–H). Arabidopsis homolog AtJMJ30 was included in parallel reactions for comparison. Unexpectedly, no 14-Dalton (Da) mass shift (indicative of methyl group removal) was detected for any of the substrates, indicating absence of demethylase activity. However, reactions containing GmJMJ19 or GmJMJ20 with H3K36me3 consistently showed a new peak at 1510 Da, corresponding to a clipping product at lysine 27 (K27) located at the C-terminus (indicated by the red arrow in Fig. 3H). Notably, GmJMJ19 and GmJMJ20 exhibited stronger activities compared to AtJMJ30, with nearly exhausted substrates, whereas approximately half of the substrate remained in the AtJMJ30 reaction (Supplementary Fig. S6A–D). Conversely, the 1510-Da peak was absent in reactions using H3K27me3 peptides, suggesting that trimethylation at K27 negatively correlates with being clipped. Given human JMJD5 cleaves arginine-methylated histone tails [9], we tested whether GmJMJ19/20 are capable of binding to and clipping H3R2me2s (symmetric), H3R8me2a (asymmetric), and H3R8me2s (symmetric) peptides. ITC confirmed binding (Supplementary Fig. S7A–C and Supplementary Table S3), however, no cleavage was detected by MALDI-TOF mass spectrometry (Supplementary Fig. S7D–F). Structural comparison with HsJMJD5 (PDB 6AX3) revealed a conserved overall fold but notable differences in the substrate-binding pocket, including substitution of the 241SRYTD245 motif with 230KNYLC234 and an S318T change (Supplementary Fig. S7G–I). These alterations likely shrink the pocket and alter side-chain orientation, preventing productive alignment for histone tail clipping of arginine-methylated substrates.
To precisely determine the clipping position and evaluate whether methylation status affects enzymatic activity, we synthesized two unmethylated H3 peptides spanning residues 1–21 and 21–44 (H3 1–21aa and H3 21–44aa). ITC assays demonstrated that GmJMJ19 binds both peptides with measurable affinity (Supplementary Fig. S8A and B), indicating recognition of multiple regions of the H3 N-terminal tail in a methylation-insensitive manner. In examining substrate specificity with the H3 1–21aa peptide, no clipping products were detected in reactions with GmJMJ19 or GmJMJ20 (Supplementary Fig. S8C). Minor peaks observed previously in H3K9me3 peptide reactions (1985 and 1857 Da, corresponding to clipping after K18 and R17; Fig. 3F) were absent, suggesting these are likely nonspecific byproducts rather than genuine enzymatic products. Enzymatic assays using the H3 21–44aa peptide revealed a consistent 1779 Da fragment generated by GmJMJ19, GmJMJ20, and AtJMJ30, indicating specific clipping at K27 (indicated by the red arrow in Supplementary Fig. S8D). Additionally, a 1691 Da fragment was uniquely observed in the AtJMJ30 reaction, likely derived from the primary 1779 Da clipping product at K27, potentially undergoing further hydrolysis due to instability, resulting in the loss of the terminal serine (Supplementary Fig. S8D). Collectively, these findings suggest that GmJMJ19/20 possess histone H3 endopeptidase activity, preferentially cleaving at unmethylated K27 residues.
Cofactor independence and structural conservation support the noncanonical function of GmJMJ19/20
The demethylation activity of canonical JmjC domain-containing proteins typically requires two cofactors: Fe^2+^ and α-KG [5, 53], which were included in the enzymatic assays as described above. Through multiple sequence alignment and AlphaFold3-based structural modeling, the putative α-KG-binding residue (K325) and Fe²⁺-binding residues (H310, D312, and H389) in GmJMJ19 were identified (Fig. 4A and Supplementary Fig. S1). To determine whether these cofactors are essential for histone-clipping activity, enzymatic assays were conducted using GmJMJ19 and the nonmethylated H3 peptide (H3 21–44aa) in the absence of either Fe²⁺ or α-KG. Notably, GmJMJ19 maintained full clipping activity without either cofactor (Fig. 4B), indicating that its endopeptidase function is cofactor-independent. The Fe-binding capacity of GmJMJ19, GmJMJ20, AtJMJ30, and the canonical JmjC demethylase GmELF6 was quantified using inductively coupled plasma−mass spectrometry (ICP−MS). As anticipated, GmELF6 exhibited the highest Fe-binding ability, while GmJMJ19, GmJMJ20, and AtJMJ30 barely bound Fe (Fig. 4C), further supporting the conclusion that metal coordination is not essential for their enzymatic function. To investigate the role of cofactor-binding residues in catalysis, key amino acids in GmJMJ19 were mutated to alanine, generating a single-point mutant (K325A) and a triple-point mutant (H310A−D312A−H389A). While both mutants cleaved at the same site as the wild type, the triple mutant showed reduced enzymatic activity (Fig. 4D). To determine whether this reduction was due to structural disruption rather than direct involvement in catalysis, we performed thermal shift assays. The K325A single mutant maintained thermal stability comparable to the wild type, whereas the triple mutant exhibited significantly decreased melting temperature (Tm), indicating protein misfolding (Fig. 4E). Taken together, these results demonstrate that the histone H3 endopeptidase activity of GmJMJ19 and GmJMJ20 operates independently of Fe²⁺ or α-KG, relying instead on protein structural integrity within the core domain.
The histone H3 endopeptidase activity of GmJMJ19 is independent of Fe²⁺ and α-KG. (A) Structural model of GmJMJ19 predicted by AlphaFold3. Putative Fe²⁺- and α-KG-binding residues are indicated. The N-terminal helical domain (Helical ND) is shown in orange, and the JmjC catalytic domain is shown in pink. (B)In vitro enzymatic assays using the H3 21–44aa peptide as substrate, comparing GmJMJ19 wild type with reactions lacking Fe²⁺ (-Fe²⁺) or α-KG (-α-KG). The complete reaction with wild-type GmJMJ19 (positive control) is shown at the top, while reactions lacking Fe²⁺ or α-KG are shown below in order. The red arrow indicates the cleavage site, while the red and blue lines represent the products and substrates, respectively. All enzymatic assays were repeated in triplicate with consistent results. (C) Fe content measurement of GmJMJ19, GmJMJ20, AtJMJ30, and GmELF6 via ICP−MS. (D) Enzymatic assays using K325A and H310A−D312A−H389A mutants. The red arrow indicates the cleavage site, while the red and blue lines represent the products and substrates, respectively. All enzymatic assays were repeated in triplicate with consistent results. (E) Thermal shift assays comparing the melting temperatures (Tm) of GmJMJ19 wild type and mutants. Derivative plots (–d(RFU)/dT versus temperature) were used to determine the Tm, corresponding to the unfolding transition midpoint. RFU represents relative fluorescence units. All assays were conducted in triplicate with independent protein preparations, and representative results are shown. (F) Structural alignment of AlphaFold3-predicted full-length models of GmJMJ19 (pink), GmJMJ20 (purple), and AtJMJ30 (yellow) with the crystal structure of HsJMJD5 (cyan, PDB 4GJZ). (G) Close-up views of the catalytic pockets of GmJMJ19, GmJMJ20, AtJMJ30, and HsJMJD5, highlighting the conserved residues that occlude substrate access. (H) Close-up views of the catalytic pockets of ceKDM7A (orange, PDB 3N9N) and HsPHF8 (green, PDB 3KV4), canonical lysine demethylases with open active sites and small side-chain amino acid residues that allow substrate entry. Residues are shown in stick representation, metal ions as spheres, and α-KG/NOG are shown in brown sticks.
Previous studies have indicated that the human homolog HsJMJD5 lacks histone demethylase activity [11]. Structural analyses revealed that HsJMJD5 (PDB 4GJZ) exhibits close structural similarity to the hydroxylase FIH-1 (PDB 2WA4), but diverges from canonical JmjC-containing lysine demethylases (KDMs). Notably, three critical residues essential for substrate binding in KDMs are substituted in HsJMJD5 (PDB 4GJZ), causing physical occlusion of the active site and preventing access to methylated lysines. The JmjC domains of GmJMJ19 and GmJMJ20 display high sequence similarity to HsJMJD5 and maintain the same substrate-occluding residues (Supplementary Fig. S1), indicating a shared structural basis for their absence of demethylase activity. To confirm this, structural superimposition was performed using AlphaFold3-predicted models of GmJMJ19, GmJMJ20, and AtJMJ30 with the experimentally determined structure of HsJMJD5. All three plant proteins exhibited high structural similarity to HsJMJD5 (Fig. 4F). Examination of the catalytic pocket revealed that in both GmJMJ19 and GmJMJ20, the same three residues (Gln266, Trp299, and Trp403 in GmJMJ19) physically obstruct substrate access, as observed in HsJMJD5 (Fig. 4G). In contrast, canonical KDMs, such as Caenorhabditis elegans KDM7A (ceKDM7A, PDB 3N9N) and human PHF8 (HsPHF8, PDB 3KV4), possess open active sites formed by small side-chain amino acid residues (Ser424, Leu484, and Asn581 in ceKDM7A), enabling efficient entry of methylated lysines (Fig. 4H). These structural characteristics strongly support the hypothesis that GmJMJ19/20 lack canonical histone demethylase activity and instead function as site-specific endopeptidases that sensitive to methylation.
GmJMJ19 and GmJMJ20 are partially redundant in regulating flowering time
The high sequence similarity and similar expression patterns between GmJMJ19 and GmJMJ20 suggest functional redundancy. To investigate the role of GmJMJ19 and GmJMJ20 in flowering time regulation, the CRISPR/Cas9 system was employed using two sgRNAs (sg1 and sg2), targeting corresponding regions in the first exon of GmJMJ19 and GmJMJ20, respectively, to generate loss-of-function mutants in the Williams 82 (W82) background (Supplementary Fig. S9A). Two independent lines from each construct were isolated and genotyped, designated as gmjmj19gmjmj20-sg1-1 and gmjmj19gmjmj20-sg2-1 (gmjmj1920-1 and gmjmj1920-2 hereafter), for subsequent analysis. In gmjmj1920-1, a 2-bp deletion at separate positions was edited for GmJMJ19 and GmJMJ20, respectively, while gmjmj1920-2 contains a 1-bp insertion in both GmJMJ19 and GmJMJ20 (Fig. 5A and Supplementary Fig. S9B and C). These mutations result in frameshifts and premature stop codons in both lines (Fig. 5B).
GmJMJ19 and GmJMJ20 regulate flowering time in SD condition. (A) Results of sequencing target sites in T1 generations in W82, gmjmj1920-1 and gmjmj1920-2. CRISPR/Cas9 editing forms and PAM sequences are shown in red and blue, respectively. (B) Predicted amino acid length of GmJMJ19 and GmJMJ20 in W82, gmjmj1920-1 and gmjmj1920-2. Red text indicates missense amino acids resulting from CRISPR/Cas9-induced frameshift mutations. (C) Flower and whole plant phenotypes of W82, gmjmj1920-1 and gmjmj1920-2 under SD condition. Both mutant lines exhibited delayed flowering compared to W82. Bar = 5 cm. (D) Flowering time analysis of W82, gmjmj1920-1 and gmjmj1920-2 shown in panel (C). P-values were calculated using the two-tailed Student’s t-test. Expression levels of GmFT2a (E) and GmFT5a (F) in W82, gmjmj1920-1 and gmjmj1920-2 at ZT4. Error bars mean standard deviations from three biological replicates. P-values were calculated using the two-tailed Student’s t-test.
To determine the in vivo cleavage activity of GmJMJ19/20 and to confirm loss of function in the mutants, we isolated histones from the leaves of W82, gmjmj1920-1 and gmjmj1920-2. Western blot analysis using antibodies against C-terminal of histone H3 and histone H4 revealed a faster-migrating band in W82, which was diminished in the gmjmj1920 mutants (Supplemental Fig. S9D). This suggests that a portion of histone H3 is cleaved in W82, and this cleavage event is compromised in the gmjmj1920.
Flowering time was examined among W82, each gmjmj19 and gmjmj20 single mutant identified from the offspring of gmjmj1920-2, and gmjmj1920-1 and gmjmj1920-2 double mutants under both SD and LD conditions. All single and double mutants exhibited delayed flowering time compared to W82 under SD conditions (Fig. 5C and D, and Supplementary Fig. S9E), with double mutants showing greater delay relative to single mutants (Supplementary Fig. S9E). However, both single and double mutants displayed flowering times indistinguishable from W82 under LD conditions (Supplementary Fig. S9F). These results indicate that GmJMJ19 and GmJMJ20 are partially redundant in regulating soybean flowering time. We further examined the expression levels of GmFT2a and GmFT5a measured at ZT4 (the expression peak) [54] in double mutants and found their significant reduction in the double mutants compared to W82 (Fig. 5E and F).
Previous studies demonstrated that enhanced expression of GmLUX1 or GmLUX2 does not influence flowering time in soybean [22]. To examine whether overexpression of either GmJMJ19 or GmJMJ20 could affect flowering time, we generated three independent transgenic soybean stable lines of GmJMJ20pro:GmJMJ20-3 × FLAG. These plants exhibited elevated levels of GmJMJ20 transcripts and its protein at various degrees (Supplementary Fig. S10A–C). Subsequently, we monitored the flowering time of the three independent lines and observed that even the highest expression of GmJMJ20 in GmJMJ20pro:GmJMJ20-3 × FLAG-1 line exhibited indistinguishable flowering time compared to W82 (Supplementary Fig. S10D). These results imply that the endogenous GmJMJ20 level may already be sufficient in vivo, and that further elevation of its expression does not alter the phenotype, consistent with previous reports showing that overexpression of GmLUX2 alone also failed to produce flowering phenotypes [22].
GmJMJ19 and GmJMJ20 promote the expression of genes involved in flowering
Since histone clipping typically affects gene transcription [55], we hypothesized that GmJMJ19/20 mutation might regulate gene expression, thereby influencing flowering. Given the direct interaction between GmJMJ19/20 and GmLUX2, we initially performed RNA-seq with high reproducibility among biological replicates on leaves of W82, lux1lux2 and gmjmj1920-2 plants (Supplementary Fig. S11A and Supplementary Table S4). In comparison to W82, gmjmj1920-2 displayed 2186 DEGs with upregulated expression and 1229 DEGs with downregulated expression, while lux1lux2 exhibited 1464 upregulated and 353 downregulated genes (Fig. 6A and Supplementary Tables S5 and S6). To identify the coregulated genes by both GmJMJ19/20 and GmLUX2, we compared the DEGs between gmjmj1920-2 and lux1lux2, and found significant overlaps with 561 upregulated and 121 downregulated genes (Fig. 6A). GO annotation of the 561 genes revealed a significant enrichment in terms related to response to stimulus and immune system process, whereas the 121 genes were predominantly enriched in processes governing growth and development (Supplementary Fig. S11B and C). To further determine the direct targets regulated by GmJMJ19/20-GmLUX1/2, we conducted ChIP followed by sequencing (ChIP-seq) using a previously validated commercial anti-AtLUX antibody in W82 [23], and anti-FLAG antibody in GmJMJ20pro:GmJMJ20-3 × FLAG-1 plants, respectively (Fig. 6B and Supplementary Fig. S12A and B). We identified 2012 reliable GmLUX binding loci globally, and genomic annotations revealed that GmLUX peaks distributed throughout the genome with predominant binding at promoters and 5′ UTR regions (Supplementary Fig. S12C). Among the top five enrichment motifs identified from these peaks, the G-box and LBS-like motifs ranked second and fifth, respectively (Supplementary Fig. S12D), both motifs have been previously reported as AtLUX binding sites [56], consistent with the EC function in suppressing gene expression. After excluding peaks annotated to distal intergenic regions, the remaining peaks were mapped to 951 genes (Supplementary Table S7). For GmJMJ20, a total of 12 267 reliable peaks associated with 6648 genes were identified (Fig. 6B, Supplementary Fig. S12B, and Supplementary Table S8). Among 178 cotarget genes of GmLUX2 and GmJMJ20, only 6 and 3 genes were recognized as common DEGs (Fig. 6C). We noticed Glyma.05G018800 among the 6 cotarget and coregulated genes of the LUX-JMJ module. In soybean, Glyma.05G018800, designated as GmFULc/GmFUL2a/Tof5/MDE05, is a member of the MADS-box gene family and contributes to flowering time regulation [27, 28, 57–59]. GmFULc funcFigtions downstream of E1 and forms a positive regulatory loop with GmFT2a/5a. GmFULc directly binds to the GmFT5a promoter to activate its transcription [27, 28, 59]. Soybean also has another three homologs of AtFUL, GmFULa (Glyma.06G205800), GmFULb (Glyma.04G159300), and GmFULd (Glyma.17G081200). Among these, GmFULd exhibited significant downregulation in both gmjmj1920-2 and lux1lux2, while GmFULa/b remained unchanged (Supplementary Table S5 and S6). GmFULa/b and GmFULc/d are the closest paralogs, respectively, with GmFULa/b and GmFULc/d diverging early [28]. Analysis of their promoter sequences revealed distinct element compositions, indicating differential regulation of GmFULa/b and GmFULc/d gene pairs (Supplementary Fig. S13A). Due to numerous degenerate bases observed in the promoter analyses of the four GmFUL genes, we examined the promoter regions between GmFULc and GmFULd, identifying well-conserved motif constitutions near the 5′ UTR (Supplementary Fig. S13B), including LBS (LUX binding sites) in motif 4 (Supplementary Fig. S13B). This was further validated by EMSA assays (Fig. 6D) and GmLUXs ChIP-seq data for GmFULc (Fig. 6E), although the signal was relatively weak at corresponding regions in GmFULd (Supplementary Fig. S14).
GmJMJ20 and GmLUX2 directly promote the GmFULc expression. (A) Venn diagram illustrating the overlap between DEGs in lux1lux2 and gmjmj1920-2. The shared DEGs coregulated by both GmLUX2 and GmJMJ19/20 are highlighted in red. (B) Metaplots of normalized signals generated from each two biological replicates of GmLUX2 and GmJMJ20 ChIP-seq. Two kilobytes genomic regions flanking the center of the peaks is shown. (C) Venn diagram showing the overlap between coregulated genes and colocated genes of GmLUX2 and GmJMJ19/20. (D) EMSA showing the binding of GmLUX2 to the probe containing LBS in the promoter of GmFULc in vitro. FAM-labeled probes were incubated with recombinant GmLUX2-SUMO or SUMO proteins. Competition assays for labeled probe were performed by adding an excess of unlabeled wild-type (cold) or mutant probes. (E) IGV screenshots displaying the normalized read counts of GmLUX2 enrichment level (anti-AtLUX ChIP-seq data), GmJMJ20 enrichment level (anti-FLAG ChIP-seq data), and H3K4me3, H3K27me3, and H3K36me3 CUT&tag assays in W82 and gmjmj1920-2 at the GmFULc locus. Y- axis value indicates normalized read counts. Gray blocks represent the associated peaks identified. (F) The schematic map presenting GmFULc promoter containing 5′ UTR. Short horizontal black lines indicate the amplicons by ChIP-qPCR. (G) ChIP-qPCR assays demonstrating the association of GmJMJ20 to GmFULc promoter. A fragment of EF1b was amplified as a control. Error bars represent the standard deviation calculated from the means of two independent biological replicates (ChIP assays), each with three technical replicates. Statistical analysis was performed using two-sided Student’s t-test.
We further validated the GmJMJ20 enrichment at the GmFULc locus through ChIP-qPCR using FLAG antibodies. We designed four primer pairs to amplify EF1b, P1, P2, and P3, among which EF1b serves as the internal reference, P1 represents the highest enrichment sites of E1 previously reported [27], and P3 corresponds to the GmLUX2 binding peak (Fig. 6F and G). It turned out that GmJMJ20 showed enrichment at the LUX binding site, with enhanced signals at P2 adjacent to the GmLUXs binding site (Fig. 6E and G). These findings support the idea that GmFULc and GmFULd are potential direct targets of GmLUXs and GmJMJ19/20 with a regulatory mechanism independent of the classic EC for gene expression suppression.
To investigated how the cleavage activity regulates gene expression, we further employed CUT&Tag to profile the levels of H3 and three key histone methylation modifications (H3K4me3, H3K27me3, and H3K36me3) in the W82 and gmjmj1920-2 mutants. CUT&Tag data showed that histone H3, levels of histone modifications across all annotated genes are indistinguishable between W82 and gmjmj1920-2 (Supplemental Fig. S15A and B), suggesting that GmJMJ19/20 do not globally affect the nucleosome landscape.
Specifically, the epigenetic status of GmFULc was altered. In gmjmj1920-2, GmFULc exhibited narrowed H3K4me3 peaks, acquired novel H3K36me3 peaks, and lost H3K27me3 peaks within its first exon and gene body (Fig. 6E). Considering that GmJMJ19/20 are unable to act on H3K27me3-marked sites and that H3K36me3 antagonizes H3K27me3 [60,61], we hypothesize that in gmjmj1920-2, the loss of GmJMJ19/20-mediated proteolytic cleavage leads to aberrant deposition on H3K36me3 disrupting the H3K4me3−H3K27me3 bivalent status of the histone tails, which ultimately suppresses GmFULc.
Natural variation of GmJMJ19 during soybean adaptation
To investigate whether GmJMJ19/20 underwent natural selection for regional adaptation, similar to GmLUX1/2 in soybean [21], we analyzed the nucleotide diversity (Fst, π, and Tajima’s D) across the 2 Mb genomic regions spanning GmJMJ19 and GmJMJ20 in a panel of 2810 accessions, including 92 wild accessions, 1020 landraces, and 1698 cultivars [62]. The analyses indicated that GmJMJ19 and GmJMJ20 are not domestication genes, based on comparisons between wild accessions and landraces, and between landrace accessions and improved cultivars (Fig. 7A–C and Supplementary Fig. S16A–C). To examine potential natural selection of GmJMJ19 and GmJMJ20, we analyzed their nucleotide polymorphisms using the same panel. Based on two SNPs located in the intron and 3′ UTR, GmJMJ19 could be classified into four haplotypes (H1–H4) (Fig. 7D), while GmJMJ20 remained conserved across all varieties. Analysis of GmJMJ19 allele frequencies revealed that H2 and H4 were predominant in wild soybeans but less frequent in landraces and cultivars, suggesting their ancestral nature. Conversely, H1 and H3 frequencies increased significantly in landraces and cultivars, indicating selection during domestication (Fig. 7E). Examination of major GmJMJ19 allele distributions among 581 Chinese landrace accessions (Supplementary Table S9) across four regions—northeast region (NE), northern region (NR), Huanghuai region (HR), and southern region (SR)—showed that while H1 and H3 maintained similar frequencies across most regions, H1 predominated in southern China (Fig. 7F). To further elucidate the function of the SNP, we examined its expression levels in three randomly selected soybean accessions that carry either H1 or H3. GmJMJ19 exhibited lower expression levels in accessions harboring H1 than those carrying H3 (Fig. 7G). This correlation suggests a potential mechanism whereby the SNP in the 3′UTR affects mRNA stability or processing, ultimately leading to differential gene expression between haplotypes. To further investigate the phenotypic effects of these two major haplotypes in landraces and cultivars, we first excluded accessions carrying two major known flowering-time genes, E1 and J, then a total of 7 and 29 near-isogenic backgrounds accessions harboring H1 and H3, respectively, were used to measure the flowering time in Guangzhou (23°16′N, 113°23′E), China. As expected, varieties carrying H1 flowered significantly later than those carrying H3 (Fig. 7H) with approximately delay 4.26 day, indicating that GmJMJ19 still produce subtle influence on flowering time under natural conditions. These findings suggest that GmJMJ19, but not GmJMJ20, underwent selection, with the H1 haplotype potentially facilitating soybean adaptation to photoperiod and environmental conditions in southern China through its role in flowering regulation.
GmJMJ19 contributes to the expansion of soybean to south China. The nucleotide variation demonstrated by Fst (A), π values (B), and Tajima’s D (C) in wild soybeans, landraces, and improved cultivars across the 2 Mb genomic regions surrounding GmJMJ19. (D) Haplotypes (H1–H4) of GmJMJ19 in genetic regions. The numbers denote nucleotide position relative to TSS of GmJMJ19 (bp). (E) Proportions of GmJMJ19 haplotypes within each of the three germplasm groups. (F) The geographical distribution of 581 landrace accessions containing H1 and H3. NE, northeast region of China; NR, northern region of China; HR, Huanghuai region of China; SR, southern region of China. (G) The expression patterns of GmJMJ19 were analyzed in representative accessions belonging to two distinct haplotypes (H1 and H3). GmACT was used as an internal control. Error bars indicate the standard deviations derived from three independent biological replicates for each haplotype. (H) Flowering time of representative H1 and H3 haplotype accessions grown under outdoor SD conditions. Five individual plants were measured for each variety.
Discussion
JmjC domain-containing proteins have been extensively studied in plants and are established as essential epigenetic regulators involved in multiple biological processes, including flowering, leaf senescence, fruit ripening, stress responses, and energy homeostasis [7, 63–66]. Traditionally, their function has been associated with histone demethylation and subsequent gene expression modulation [67]. However, recent evidence indicates that certain JMJ proteins operate through demethylase-independent mechanisms. For instance, AtJMJ30 does not directly regulate H3K36 methylation at circadian loci [68], and AtJMJ30 promotes H3K36me3 deposition at LATERAL ORGAN BOUNDARIES-DOMAIN (LBD) genes by recruiting ARABIDOPSIS TRITHORAX-RELATED 2 (ATXR2) [69], while AtJMJ24, AtJMJ26 and AtJMJ28 function as subunits of COMPASS (Complex of Proteins Associated with Set1) complexes [70], facilitating interactions with the chromatin remodeler INOSITOL REQUIRING80 (INO80) to activate transcription via H3K4me3 modification [70].
Notably, we demonstrate that AtJMJ30, previously characterized as demethylase targeting H3K36me3, H3K27me3, and H3K9me3 during flowering and callus formation [69, 71, 72], exhibits endopeptidase activity as well, specifically cleaving histone H3 at unmethylated lysine 27 (H3K27) (Supplementary Fig. S6A–D). This finding contradicts previous studies that inferred demethylase activity from reduced methylation levels in vivo, primarily through ChIP-seq or Western blotting. These methods cannot definitively distinguish direct demethylation from indirect effects, such as histone tail clipping or the recruiting of other demethylases by AtJMJ30. Our in vitro MALDI-TOF assays, detected no 14-Da demethylation shift but a 1779-Da clipping product, establish the endopeptidase activity of AtJMJ30. This activity parallels that of JMJD5 and JMJD7, which cleave H3 tails near methylated arginine residues under stress conditions [9, 10, 16]. Similarly, the soybean homologous pair GmJMJ19 and GmJMJ20 exhibit histone H3 endopeptidase activity, preferentially cleaving at unmethylated K27 rather than arginine-methylated sites (Fig. 3E–H and Supplementary Fig. S7D–F). This difference between GmJMJ19/20 and JMJD5 in substrate preference, likely arising from subtle alterations in the substrate-binding pocket, such as residue substitutions reducing pocket size and altering side-chain orientation, highlights evolutionary divergence within the JmjC family. These observations, combined with structural analyses showing that HsJMJD5 shares high similarity with hydroxylases like FIH, support the enzymatic versatility and mechanistic divergence of JmjC proteins [15, 73]. Nevertheless, it remains possible that JMJ proteins lacking demethylase activity may interact with other histone demethylases to indirectly alter histone codes and regulate gene expressionin vivo.
GmJMJ19/20 interact with GmLUX2 to promote GmFULc, a MADS-box gene essential for SD flowering (Fig. 6D–G). The gmjmj1920 mutants display delayed flowering and decreased GmFT2a/5a expression (Fig. 5C–F), suggesting that GmJMJ19/20′s clipping activity, potentially facilitated by GmLUX2-mediated recruitment to GmFULc and other loci, enhances flowering. RNA-seq and ChIP-seq data demonstrate that GmJMJ19/20 and GmLUX2 coregulate a subset of flowering genes, with GmFULc and GmFULd as direct targets (Fig. 6C–G, Supplementary Fig. S14, and Supplementary Tables S5 and S6). While GmJMJ19/20 were previously associated with demethylation at the GmZF351 locus under salt/mannitol stress [74], our findings indicate their primary role in flowering involves histone clipping rather than demethylation, emphasizing the context-dependent functional plasticity of JMJ proteins. These results expand the functional understanding of plant JMJs, establishing histone clipping as a novel mechanism in flowering regulation.
Here, we present a model illustrating how LUXs associate with JMJ19/20 to mediate flowering time in soybean (Fig. 8). In wild type, LUXs typically form EC with ELF3a and ELF4a/4b, similar to their Arabidopsis counterparts [22, 23, 75]. The EC represses the downstream gene E1 expression [22, 23], likely through establishing repressive chromatin states such as H3K27me3, as documented in Arabidopsis [76–81]. E1 functions to repress the expression of GmFT2a/5a, thereby mediating flowering time (Fig. 8A). Beyond direct binding of E1 by EC, recent studies have identified an alternative pathway wherein E1 represses GmFT2a/5a through inhibition of GmFULc expression [27, 59]. These findings support the existence of a more complex regulatory network containing E1, GmFULc and GmFT2a/5a. Notably, our research provides evidence that GmLUXs directly promote GmFULc expression through interaction with GmJMJ19/20 in both histone demethylase activity- and EC-independent manners. This conclusion is supported by several findings: GmLUXs interact with GmJMJ19/20 both in vitro and in vivo; GmJMJ19/20 mutation results in late flowering; GmJMJ19/20 possess histone H3 endopeptidase activity but lack histone demethylase activity; and GmLUXs and GmJMJ20 directly associate with the GmFULc promoter to promote its expression independently of EC. Thus, GmFULc expression positively regulated by GmLUXs occurs through two mechanisms: inhibition of E1 to release its expression, and association with GmJMJ19/20 to promote its expression. In the gmjmj1920 double mutant, reduced GmFULc levels subsequently downregulate GmFT2a/5a, resulting in delayed flowering time (Fig. 8B).
A working model showing the role of GmJMJ19/20 in regulating flowering time in soybean. (A) The soybean EC, comprising GmELF3s, GmLUXs, and GmELF4s, directly suppresses the expression of E1 family members, thereby release the repression of GmFT2a/5a to promote flowering time. The right portion of the dashed box illustrates that LUXs also operate through EC-independent mechanisms in certain regions, such as interacting with GmJMJ19/20 presumably by clipping unmethylated H3K27 to establish a relaxed chromatin state, thus enhancing GmFULc expression for flowering time regulation. (B) When GmJMJ19/20 is absent, the chromatin structure at target sites becomes relatively compact, negatively affecting gene expression, including GmFULc, consequently delaying flowering time. Genes marked in blue indicate expression levels lower than W82.
It has been shown that complete knockout of GmLUX1/2 lost photoperiod sensitivity with delayed flowering under both SD and LD [22], while the gmjmj1920 mutant exhibits a photoperiod-specific phenotype only under SD. This difference suggested that LUXs may function together with JMJs in SD in an EC-independent manner. Here, we propose that LUXs associate with ELF3/4 to form the EC suppressing transcription, but promote expression of genes including FULc through JMJs. Since GmFULc reportedly expresses predominantly under SD conditions and promotes flowering via triggering the expression of GmZTL3 and GmZTL4 genes [28], the LUX-JMJ-FULc module may function specifically under SD in an EC-independent manner.
An intriguing aspect is the relationship between GmLUXs and GmJMJ19/20. We noticed that GmJMJ20 exhibits significant downregulation (log_2_FC = −1.75, *P-*value = 5.93e−05) in lux1lux2, and GmJMJ19 shows moderate downregulation (log_2_FC = −1.37, *P-*value = .0031), but not vice versa (Supplementary Tables S5 and S6). Though we do not have direct evidence at this moment, this suggests that GmLUX2 may trigger or enhance the rhythmic expression of GmJMJ19/20 and also recruit GmJMJ19/20 to modulate gene expression through histone clipping at specific loci. Whether this regulation is dependent on EC or not, GmLUXs and GmJMJ19/20 form a feed-forward loop in gene expression regulation.
Flowering time is an essential agronomic trait in soybean due to its photoperiod sensitivity. Research in Arabidopsis and rice demonstrates that flowering time regulation incorporates both epigenetic and photoperiodic pathways [82]. However, epigenetic regulators in soybean remain largely unexplored, potentially due to its genomic complexity hindering efficient mutant screening. Recent research identified GmLDL2 as a regulator of flowering time through repression of GmFER expression via modification of H3K4 methylation status at its gene locus [83]. During domestication, GmLDL2 underwent both natural and artificial selection, with two haplotypes selected for different latitudes. In this study, our findings not only identify JMJ proteins with histone H3 endopeptidase activity but also elucidate their function in flowering time control. Thus, the epigenetic regulation on soybean flowering is becoming evident, and will advance our understanding of fine-tuned flowering time regulation under natural environments in soybean.
Supplementary Material
gkag134_Supplemental_Files
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1Bannister AJ, Kouzarides T. Regulation of chromatin by histone modifications. Cell Res. 2011;21:381–95. 10.1038/cr.2011.22.21321607 PMC 3193420 · doi ↗ · pubmed ↗
- 2Kouzarides T . Chromatin modifications and their function. Cell. 2007;128:693–705. 10.1016/j.cell.2007.02.005.17320507 · doi ↗ · pubmed ↗
- 3Berger SL . The complex language of chromatin regulation during transcription. Nature. 2007;447:407–12. 10.1038/nature 05915.17522673 · doi ↗ · pubmed ↗
- 4Liu C, Lu F, Cui X et al. Histone methylation in higher plants. Annu Rev Plant Biol. 2010;61:395–420. 10.1146/annurev.arplant.043008.091939.20192747 · doi ↗ · pubmed ↗
- 5Klose RJ, Kallin EM, Zhang Y. Jmj C-domain-containing proteins and histone demethylation. Nat Rev Genet. 2006;7:715–27. 10.1038/nrg 1945.16983801 · doi ↗ · pubmed ↗
- 6Lu F, Li G, Cui X et al. Comparative analysis of Jmj C domain-containing proteins reveals the potential histone demethylases in Arabidopsis and rice. JIPB. 2008;50:886–96. 10.1111/j.1744-7909.2008.00692.x.18713399 · doi ↗ · pubmed ↗
- 7Li Z, Jiang G, Liu X et al. Histone demethylase Sl JMJ 6 promotes fruit ripening by removing H 3K 27 methylation of ripening-related genes in tomato. New Phytol. 2020;227:1138–56. 10.1111/nph.16590.32255501 · doi ↗ · pubmed ↗
- 8Qian S, Wang Y, Ma H et al. Expansion and functional divergence of Jumonji C-containing histone demethylases: significance of duplications in ancestral angiosperms and vertebrates. Plant Physiol. 2015;168:1321–37. 10.1104/pp.15.00520.26059336 PMC 4528752 · doi ↗ · pubmed ↗