The Arabidopsis thaliana ecotype Columbia-0 (Col-0) or mutants in the Col-0 background were used in all experiments. The T-DNA insertion lines GABI_355H03 (hd2a), Sail_1247_A02 (hd2b), Salk_039784 (hd2c), GK_379G06 (hd2d), and SM_3_20886 (dog1-4) were described previously (Luo et al., 2012a; Fedak et al., 2016; Li et al., 2017) and were verified by PCR on genomic DNA using gene-specific primers ( Supplemental Table S1 ). The double mutant hd2ahd2b, hd2ahd2c, hd2ahd2d, hd2bhd2c, hd2bhd2d, hd2chd2d, and the triple mutants hd2ahd2bdog1-4 were produced by crossing. Homozygous lines were isolated by genotyping with gene-specific primers ( Supplemental Table S1 ). The HD2A and HD2B complementation lines were gifts from Ton Bisseling (Li et al., 2017). Seeds were sown in moist soil mixed with sand in a ratio of 10:1 and cultivated in the growth chambers under long-day conditions (14 h light/10 h dark and 20°C/18°C, respectively) or short-day conditions (10 h light/14 h dark and 20°C/16°C, respectively). The Arabidopsis plants used for seed production were grown first under short-day conditions for 4 weeks before being transferred to long-day conditions for flowering. The seeds were harvested and stored in the dark at room temperature.

Measurements of HDA activity of Arabidopsis tissue protein extracts were performed by a fluorescence-based method adapted from Wegener et al. (Wegener et al., 2003a; Wegener et al., 2003b) and (Nott et al., 2008). 150 mg of ground deep-frozen plant material per replicate was transferred to a pre-cooled Lysing Matrix D tube (MP Biomedicals, Santa Ana, California, USA) and homogenized for 1 min at full speed in a FastPrep^®-24 homogeniser (MP Biomedicals). After placing the tubes on ice, 300 µl homogenization buffer (50 mM Tris-HCl pH 7.0, 1 M D-Glucose, and 1x protease inhibitor cocktail) was added and the samples were again homogenized for 30 sec. The supernatant was transferred to 1.5 ml microcentrifuge tubes and centrifuged for 10 min at 25,000g and 4°C to remove cell debris. The protein concentration of the supernatant was determined according to Bradford (Bradford, 1976) and adjusted to a concentration of 1.2 µg/µl with homogenization buffer. HDA activity of the supernatant was assayed in 30 µl fractions per replicate in a flat-bottom 96-well black microtiter plate. 100 µM BOC-(acetyl) Lys-AMC (Bachem, Bubendorf, Switzerland) in 25 µl HDA reaction buffer (25 mM Tris-HCl, pH 8.0, 137 mM NaCl, 2.7 mM KCl and 1 mM MgCl₂) was added. After incubation for 2 h at 37°C, 10 mg/ml trypsin and 1 µ M TSA in 60 µl, HDA stopping buffer (50 mM Tris-HCl, pH 8.0, 100 mM NaCl) was added per well and fluorescence output, representing HDA activity, was measured after 20 min incubation at 30°C at λ 380nm_Excitation and 440nm_Emission.

For Co-IP analysis, the 35Spro : HSI2-myc and 35Spro : HSL1-myc constructs were transiently expressed in 35Spro : HD2A-GFP and 35Spro : HD2B-GFP protoplast. After brief centrifugation (100g, RT, 3 min), the supernatant was removed and total protein was extracted by re-suspending and disrupting the protoplast with 1 ml of extraction buffer (50 mM Tris-HCl, pH 8, 1 mM PMSF, 5% glycerol, 150 mM NaCl, 0.1% Nonidet P-40, 5 mM MgCl₂, 1 mM DTT, and 1x protease inhibitor cocktail). After gentle shaking for 1 h at 4°C, the sample was centrifugated at 12,000g for 10 min. To purify GFP-tagged proteins, the supernatant was incubated with 25 μl GFP-Trap agarose beads (Nano tag; catalog no. N0510) at 4°C overnight by gentle rotation. After washing with extraction buffer four times, proteins were eluted with 50 μl 2x SDS sample buffer and analyzed by immunoblotting using an anti-Myc antibody.

The ChIP-qPCR assay was performed as previously described (Bowler et al., 2004). The chromatin was extracted from 24 h imbibed WT and hd2ahd2b seeds and from 10 d old 35Spro: HD2B-GFP seedlings. The seeds were imbibed at room temperature under the dark, and at that time point, no germination was visible. About 1 g of imbibed seeds and 2 g of seedlings were cross-linked in cross-linking buffer (400 mM sucrose, 10 mM Tris-HCl pH 8.0, 5 mM β-mercaptoethanol, 1% formaldehyde) by vacuum infiltration for 1 h and 10 min, respectively. Cross-linking was stopped by adding glycine to an end concentration of 0.125 M and additional vacuum infiltration for 5 min. The cross-linked plant materials were washed two times with ice water, dried with paper towels, and ground in liquid nitrogen to a fine powder. The chromatin was extracted with 20 ml extraction buffer (400 mM sucrose, 10 mM Tris-HCl pH 8.0, 5 mM β-mercaptoethanol, 1x protease inhibitor cocktail) by gentle shaking for 20 min and pelleted by centrifugation at 4000g for 25 min at 4°C. The pellet was washed with nuclei washing buffer (20 mM Tris/HCl, pH 7.4, 25% glycerol, 2.5 mM MgCl₂, 0.2% Triton x-100), re-suspended in 600 μl nuclei sonication buffer (50 mM Tris-HCl pH 8.0, 10 mM EDTA pH 8.0, 1% SDS, 1x protease inhibitor cocktail) and the chromatin was sheared to 200–1,000 bp by sonication. the nuclei were transferred into 1.5 ml Bioruptor Microtubes (Cat No. C30010016), and 15 cycles with 30 sec ON/OFF was used with Bioruptor^® Pico ultrasonic bath and Covaris E220 Evolution. After centrifugation for 10 min at 12,000g and 4°C the supernatant was directly used for immunoprecipitation with specific antibodies. For H3K9ac, H4K5ac, and H4ac analysis, the antibodies (anti-H4ac, anti-H4K5ac, anti-H3K9ac) were coupled to the magnetic protein G beads by incubating at 4°C on a rotation platform overnight. Afterward, 100 μl sonicated chromatin was mixed with antibody-magnetic protein G beads and incubated overnight at 4°C on a rotating platform. The beads were sequential washed with low salt buffer (0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl pH 8.0, 150 mM NaCl), high salt buffer (0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl pH 8.0, 500 mM NaCl), LiCl buffer (0.25 M LiCl, 1% NP-40, 1% sodium deoxychlorate, 1 mM EDTA, 10 mM Tris-HCl pH 8.0), and TE buffer (10 mM Tris-HCl pH 8.0, 1 mM EDTA). And the beads were washed twice with each buffer. Finally, the chromatin was eluted with 400 μl of elution buffer (1% SDS, 100 mM NaHCO₃) by incubation for 20 min at 65°C, and the chromatin de-crosslinking was performed at 65°C for over 6 h after adding 16 μl of 5 M NaCl. After being treated with proteinase K and RNase, DNA was purified by phenol-chloroform method, eluted with dH₂O, and quantified for qPCR. For anti-GFP analysis, the GFP-Trap agarose beads were used instead of antibody-coupled magnetic protein G beads, and analysis was performed with the same procedure as above.

For bimolecular fluorescence complementation assays, the ORF of HD2A, HD2B, HSI2, and HSL1 (without stop codon) were transferred into the pDONOR221 vector by Gateway Cloning and subsequently shifted into the pBiFCt-2in1-NN vector (LR reactions) according to the described (Grefen and Blatt, 2012). Then, the constructs were transferred into Arabidopsis protoplasts by PEG transformation as described above. After incubation for 16 h to 24 h in the dark, the YFP fluorescence signal was monitored using a laser scanning confocal microscope (Leica TCS SP8 confocal).

The endogenous ABA and GA3 contents were measured with Agilent 1290 Infinity II-6470 triple quadrupole LC/MS/MS System according to the previously reported method (Liu et al., 2021) with minor modifications. Briefly, 0.1 g dry seeds and 0.3 g 24 h imbibed seeds were ground into a fine powder with liquid nitrogen and were transferred into a 2 ml microtube containing 1 ml ethyl acetate. The samples were vortexed and incubated for 30 min at 4°C on a shaker. Afterward, the samples were centrifuged at 12,000g for 10 min at 4°C and the supernatant was transferred into a 1.5 ml tube and dried (speed vac). The residue was re-dissolved in 200 μl of 50% methanol and filtered through a 0.22 μm filter for sample loading. For each sample, 100 μl methanol solution was subjected to LC-MS/MS analysis. ABA (yuan ye biotech, catalog no. B50724) and GA3 (yuan ye biotech, catalog no. B20187) were used as authentic reference standards. All determinations were performed in triplicate.

To investigate the precise functions of HD2s in Arabidopsis germination, we analyzed seed germination of all four HD2s T-DNA insertion lines of Col-0 background under long-day conditions, designated as hd2a, hd2b, hd2c, and hd2d ( Figure 1A ). Reduced transcript accumulation in 24h-imbibed seeds was confirmed by qRT-PCR. The results showed that nearly no transcripts of HD2A, HD2C, and HD2D were detectable ( Figure 1B ), whereas transcription of HD2B is reduced by approx. 80% ( Figure 1B ). Moreover, double mutants were generated by crossing the HD2 T-DNA insertion lines to determine, if functional redundancy is existing in seed germination among the different HD2s. Freshly harvested seeds were used for seed dormancy analysis. After three days of incubation, between 80% and 85% of WT, hd2a, hd2c, and hd2d seeds were germinated, whereas, only 70% of hd2b seeds germinated ( Figure 1C ). Among the six double mutant lines, hd2ahd2b double mutant showed a strongly enhanced seed dormancy phenotype, only around 15% germinated ( Figure 1C , Supplemental Figure S1 ). These results suggest that HD2A is at least partly functional redundancy to HD2B in activating seed germination. After 4 weeks of dry storage at 4°C, seeds of WT and hd2a, and hd2b single mutant lines germinated almost 100%, whereas hd2ahd2b seeds germinated only to 25% and even after extended storage of 16 weeks only to 65% ( Figure 1D ). These results indicated that HD2A and HD2B play an important role in promoting seed germination.

To unveil the underlying function of HD2A and HD2B in seed dormancy release, their temporal expression pattern during seed maturation and imbibition was examined by RT-qPCR. HD2A has an analogous expression pattern as HD2B, rapidly increasing from 12 d after pollination (DAP) and reaching the highest expression level in 12 h imbibed seeds ( Figure 2 ). In stored seeds, the expression level of HD2B is significantly higher than that of HD2A. In general, imbibed seeds displayed significantly higher expression levels of both genes than maturating seeds ( Figure 2 ). These results imply a function of HD2A and HD2B in seed dormancy establishment as well as dormancy release.

To analyze, if HD2A and HD2B regulate seed germination via ABA and/or GA biosynthesis and signal transduction pathways, transcripts of genes involved in ABA and GA3 biosynthesis/catabolism/signaling have been quantified ( Figures 3A, B ). The transcription of ABA1 and CYP707A2, involved in ABA biosynthesis and catabolism, respectively, was increased in hd2ahd2b ( Figure 3A ), whereas the expression of NCED3, SnRK2.3, and ABI2, which are related to ABA synthesis and ABA signal transduction, was not significantly different between hd2ahd2b and WT ( Figure 3A ). In contrast, expression of ABI5, another ABA signal transduction-related gene, was enhanced in hd2ahd2b ( Figure 3A ).

Regarding genes involved in GA biosynthesis, the expression of GA3OX1 but not that of GA3OX2 was upregulated in the hd2ahd2b seeds in comparison to WT. Moreover, the expression of GA2OX2, a gene required for GA catabolism, was not changed ( Figure 3B ). Furthermore, the analysis of the GA-repressed gene WRI1 and the gibberellin receptors encoding genes GID1b and GID1c revealed that GID1b was downregulated in hd2ahd2b, whereas the expression of WRI1 and GID1c was not significantly different in WT and hd2ahd2b seeds ( Figure 3B ).

Since the expression of at least a few genes related to ABA and GA3 biosynthesis/catabolism/signaling is affected in hd2ahd2b, we determined the content of ABA and GA3 in WT and hd2ahd2b seeds. ABA and GA3 content is lower in 24 h imbibed seeds in comparison to dry seeds of WT and hd2ahd2b plants ( Figures 3C, D ). Surprisingly, the amount of ABA and GA3 was not significantly different neither in dry nor in imbibed seeds of hd2ahd2b and WT ( Figures 3C, D ).

In conclusion, although ABA and GA3 content is not significantly affected in hd2ahd2b dry and imbibed seeds, the upregulation of ABI5 indicated that HD2A and HD2B somehow could function in the ABA signaling pathway to induce seed dormancy.

Since mutants with a seed dormancy phenotype are usually hypersensitive to ABA (Zhao et al., 2015; Née et al., 2017), we analyzed the germination of fully after-ripened hd2ahd2b and WT seeds in presence of different concentrations of ABA. hd2ahd2b displayed significantly reduced seed germination with increasing concentrations of ABA, whereas no effect was observed in seed germination of WT and the single mutant lines, demonstrating that only the double mutant hd2ahd2b is hypersensitive to ABA ( Figure 4A ). A time course experiment demonstrated that hd2ahd2b seeds are 100% viable, but showed delayed germination already in absence of ABA ( Supplemental Figure S2 ). Moreover, we tested germination of hd2ahd2b in presence of 100 µM of GA3 and after stratification at 4°C for 3 days. Both treatments slightly promoted the germination of freshly harvested and completely ripened hd2ahd2b seeds ( Figure 4B ).

Besides ABA, the protein DELAY OF GERMINATION 1 (DOG1) is an essential regulator of seed dormancy (Bentsink et al., 2006; Nakabayashi et al., 2012; Huo and Bradford, 2016). The gradually elevated expression of HD2A and HD2B during seed maturation and the fact that the hd2ahd2b double mutant line displays a “hypersensitive to ABA” germination phenotype, let us assume that HD2A and HD2B affect the expression of DOG1. Therefore, we analyze the relative expression level of DOG1 in imbibed seeds of WT, the single mutant lines hd2a and hd2b, the double mutant line hd2ahd2b, the double mutant line either complemented with HD2A-GFP (pHD2A: HD2A-GFP) or HD2B-GFP (pHD2B: HD2B-GFP) and the two HD2B overexpression lines HD2B-OE9, HD2B-OE14. DOG1 transcription level in the pHD2B: HD2B-GFP line was comparable to that of WT seeds ( Figure 5A ). In contrast, the DOG1 expression level was significantly decreased in the HD2B overexpression lines and elevated in the line with a lower HD2B transcription level ( Figure 5A ). Notably, in seeds of the hd2ahd2b double mutant, the DOG1 expression level was increased more than 20 times in comparison to WT, while in hd2a and hd2b expression of DOG1 was only two- and six-fold increased, respectively ( Figure 5A ). Surprisingly, the expression level of DOG1 in the pHD2A: HD2A-GFP complementation line is significantly higher than in the WT ( Figure 5A ). In conclusion, higher HD2B expression correlates with lower DOG1 expression indicating that HD2B may repress DOG1 expression during seed germination. Moreover, both, HD2A and HD2B functions are essential for regulating the expression of DOG1 during seed germination.

We further analyzed the dynamics of DOG1 expression in hd2ahd2b during seed maturation and imbibition. We observed that the DOG1 mRNA accumulation decreased from 12 DAP until seed maturation, which is consistent with the reported data in the Col-0 background (Zhao et al., 2015). Then, DOG1 expression increased rapidly in dry seeds and quickly vanished after seed imbibition ( Figure 5B ). Different from the expression pattern in WT, the DOG1 transcript level in the hd2ahd2b double mutant increased during seed maturation, peaked in dry seed, and decreased during imbibition. Interestingly, after an initial significant decrease at beginning of imbibition, DOG1 expression in hd2ahd2b seeds did not vanish as observed in WT seeds but remained at a relatively stable level over at least 24 h ( Figure 5B ). In general, the DOG1 expression level in hd2ahd2b is significantly higher during seed development and imbibition in comparison to WT ( Figure 5B ), concluding that freshly harvested seeds of the hd2ahd2b line might accumulate more DOG1 than WT seeds. To further get evidence for a coordinated function of HD2A/HD2B and DOG1 in seed germination, we analyzed seed germination of Arabidopsis lines with different HD2A/HD2B and DOG1 expression levels. Compared to all other lines analyzed, the seeds of the hd2ahd2b double mutant with the highest DOG1 mRNA level showed significantly lower (delayed) germination. Interestingly, this reduced germination phenotype is restored in the complementation line pHD2B: HD2B-GFP and partially in pHD2A: HD2A-GFP. Although the dog1 mutant and the HD2B overexpression lines HD2B-OX9 and HD2B-OX14 showed reduced expression of DOG1 and a similar percentage of germination as WT seeds after 2 days of incubation, the percentage of germination of WT is slightly delayed in comparison to that of dog1 and both HD2B overexpression lines ( Figure 5C ).

To demonstrate the functional relationship between HD2A, HD2B, and DOG1, we crossed hd2ahd2b and dog1-3 and dog1-4 to generate hd2ahd2bdog1-3 and hd2ahd2bdog1-4 homozygous plants. Almost all triple mutants’ seeds germinated after incubation for 48 h, indicating a none dormancy phenotype similar to dog1 mutants ( Figure 5C ). In conclusion, the DOG1 function is responsible for hd2ahd2b mediated seed dormancy. All these results suggested that HD2A and HD2B may promote seed germination by inhibiting DOG1 expression.

HD2A and HD2B are annotated as HDAs, however, their exact biochemical functions in Arabidopsis are still unknown. To confirm that HD2A and HD2B are required for histone deacetylation, the relative HDA activity in 10 days old seedlings of hd2ahd2b and WT plant was measured. hd2ahd2b displayed a 45% reduction in total HDA activity compared to WT ( Figure 6A ), suggesting that HD2A and HD2B are required for HDA activity in Arabidopsis.

Furthermore, the global acetylation levels of H4, H4K5, and H3K9 in germinated seeds of WT and hd2ahd2b double mutant were analyzed. In hd2ahd2b a 1.4-fold and 2.2-fold enhanced H4ac and H4K5ac level, respectively, was detected, but no significant difference was observed in the H3K9ac level ( Figures 6B, C ). Subsequently, we analyzed whether the acetylation level at the DOG1 promoter and the DOG1 coding region is altered in the hd2ahd2b line in comparison to WT. We performed chromatin immunoprecipitation quantitative PCR (ChIP-qPCR) on 24 h imbibed seeds of WT and hd2ahd2b using specific anti-H3K9ac, anti-H4ac, and anti-H4K5ac antibodies. A 103 bp promoter region P1, 1064 bp upstream of the transcription start site (TSS), and a 189 bp coding region P2, 157 bp downstream of TSS, were amplified with specific primers ( Figure 7A ). Loss of HD2A and HD2B function enhanced the acetylation level of H4ac and H4K5ac at the coding region P2 of DOG1, but not at the promoter region P1 ( Figure 7B ). In contrast, the H3K9ac level did not significantly change in hd2ahd2b in comparison to WT, neither in the DOG1 promoter region P1 nor the DOG1 coding region P2 ( Figure 7B ). To check, whether HD2B directly binds to the DOG1 locus, we performed ChIP-qPCR on 24 h imbibed seeds of WT and hd2ahd2b complemented with 35Spro: HD2B-GFP using an anti-GFP antibody. Arabidopsis intergenic region between AT5G43175 and AT5G43180 was selected as the negative control. Fragments corresponding to the DOG1 coding region P2 were significantly enriched concluding that HD2B-GFP binds to DOG1. No enrichment was detected in the promoter region P1 and the negative control. In sum, these data indicate that the up-regulation of DOG1 in hd2ahd2b is at least partly due to increased histone acetylation at the DOG1 coding region P2.

HD2A and HD2B need to be recruited to the different DNA binding sites by different transcriptional regulators and form multi-protein complexes to modulate chromatin structure and consequently regulate gene expression in the different development stages (Liu et al., 2014). HSI2 and its homolog HSI2-like 1 (HSL1), also known as VAL1 and VAL2, respectively, repress DOG1 expression by recruiting LIKE HETERCHROMATIN PROTEIN 1 (LHP1) and CURLY LEAF (CLF) for consequent deposition of H3K27me3 marks at DOG1 locus (Chen et al., 2020).

We assumed that HD2A and HD2B are also recruited to DOG1 by HSL1 and HSI2. The interaction between HD2A and HD2B was demonstrated by bimolecular fluorescence complementation (BIFC) ( Figure 8A ). To analyse the interaction of HD2A/HD2B with HSL1/HSI2, HD2A, and HD2B were fused to the N-terminus of YFP, and HSL1 and HSI2 were fused to the C-terminus of YFP. YFP signals were observed in the nucleus whenever HD2A or HD2B were co-expressed with HSL1 or HSI2 ( Figure 8B ). No signals were detected when HD2A and HD2B were co-transfected with the empty plasmid containing YC-YFP (negative control, Figure 8B ). Co-IP confirmed the interactions observed in the BIFC assay, since both, HSI2 and HSL1, co-immunoprecipitated with HD2A-GFP and HD2B-GFP ( Figure 8C ). These results indicate that HD2A and HD2B could be recruited by HSI2 and HSL1 to the DOG1 locus.

Seed development comprises embryo morphogenesis and seed maturation (Baud et al., 2008). The phenotype of Arabidopsis cotyledons was determined during embryo morphogenesis and the maturation process ensures the embryo accumulates enough storage reserves, which are important for seed dormancy and desiccation tolerance establishment (Baud et al., 2008; Carrillo-Barral et al., 2020). The abnormal development of the seeds plays important role in seed dormancy (Focks and Benning, 1998; Debeaujon et al., 2018), and embryo dormancy and coat-imposed dormancy are the two major types of seed dormancy mechanisms (Bewley, 1997).

Here we first found the displayed dysplastic cotyledons of hd2ahd2b seedling (tri-cotyledony, fused cotyledons, asymmetric cotyledons, and in the most extreme cases blurred border or junction between the petiole and the blade) ( Figure 9I ), which indicating abnormal embryo morphogenesis. Furthermore, the Arabidopsis WT and hd2ahd2b seeds were phenotyped using the phenoSeeder (Jahnke et al., 2016), which consists of a pick-and-place robot and several sensors, enabling measurement of seed traits such as mass, volume, density, length, width, and size (i.e., projected area) for individual seeds. Although hd2ahd2b seeds have an irregular surface, there were no relevant differences between mean traits of the genotypes ( Figures 9A–G ; Table 1 ), except that trait distributions of hd2ahd2b were wider and showed deviations from normal distributions for seed mass and volume ( Figures 9E, F ) (Bourque et al., 2011). We observed smooth testa, oval-shape, and brown color for WT seeds, whereas hd2ahd2b seeds displayed abnormal seed phenotype, with a wrinkled epidermis, irregular shape, and deeper testa color ( Figures 9G, H ). All those seed phenotypes are similar to the phenotype of mutants in seed maturation regulators and seed coat mutants, which could affect seed germination (Focks and Benning, 1998).

It was reported that DOG1 was involved in seed development by affecting multiple aspects of seed maturation via genetic interaction with ABI3 (Dekkers et al., 2016). More important, The abnormal phenotypes of hd2ahd2b largely were recovered in hd2ahd2bdog1 triple mutants ( Figure 9I ), suggesting a genetic interaction between DOG1 and HD2A/HD2B. These findings demonstrate a possible underlying dormancy mechanism caused by the up-regulation of DOG1 in hd2ahd2b lines, that HD2A and HD2B are involved in seed dormancy partly by regulating seed development.

Our data suggest that HD2A and HD2B have important function in seed germination and seedling development. To get a general overview of the physiological processes, HD2A and HD2B are involved in, we performed an RNA-sequencing (RNA-seq) analysis of ten days old WT and hd2ahd2b seedlings and 24h imbibed seeds. Differentially expressed genes were defined based on a threshold of at least 2-fold change (P-value < 0.05). In seedlings, we found, that 1720 and 772 genes were up- and down-regulated in hd2ahd2b, respectively ( Figure 10A ; Supplemental Table S2 ), demonstrating the repressive function of plant-specific histone deacetylases (Wu et al., 2003; Zhou et al., 2004; Li et al., 2017; Chen et al., 2018). Interestingly, approx. 45% of the up-regulated genes (~775), but only 15% of the down-regulated genes (~120) have a function related to “response to stimuli” ( Figure 10B ). Moreover, significantly more genes related to “localization” and “growth and development” are up-regulated in hd2ahd2b seedlings ( Figure 10B ). The Gene Ontology enrichment analyses of 2492 differentially expressed genes revealed that within the up-regulated genes, genes related to “response to different chemicals/stimuli”, “seed dormancy” and “seed maturation” were highly enriched, whereas within the down-regulated genes, genes related to “sugar and sulfur metabolic processes” were enriched ( Figure 10C ). Interestingly, DOG1-like 1 (At4g18660) and DOG1-like 3 (At4g18690) genes are up-regulated in hd2ahd2b seedlings, further confirming the repressive function of HD2A and HD2B on DOG1 gene family. Furthermore, transcription levels of genes responding to ABA and GA were also affected. These results demonstrate that HD2A and HD2B function is required for shutting down stimuli responses and genes involved in seed development and germination processes in ten days old seedlings.

In 24h imbibed seed, more genes are down-regulated (~1770) in hd2ahd2b than up-regulated (~520) ( Supplemental Figure S3 ; Supplemental Table S3 ). GO term enrichment analysis revealed that, HD2A and HD2B are also involved in regulation of genes related to “response to stress and stimulus”. Moreover, the function of both plant-specific histone deacetylases is required for regulation of genes involved in “response to ABA”, “seed development”, “post-embryonic development”, and other developmental processes ( Supplemental Figure S3 ). These results further confirm, that HD2A and HD2B function is required for regulating seed germination and further seedling development.

A previous study showed that HD2A and HD2C have contrasting roles in seed germination through glucose signaling, where HD2A restrains germination and HD2C promotes germination (Colville et al., 2011). In another study, HD2A function positively correlated with seed germination and negatively with dormancy-associated genes (Footitt et al., 2015). Similar to the later report, our results provide evidence that HD2A positively affects seed germination since HD2A single ko-mutant has an elevated DOG1 expression level ( Figure 5A ). Surprisingly, we did not observe a significant difference in the seed germination phenotype between hd2a and WT ( Figure 1D ). Probably two-fold upregulation of DOG1 in hd2a is insufficient for a detectable delay of germination. In contrast, HD2B knock-down resulted in a seven-fold upregulation of DOG1 and significantly enhanced dormancy ( Figures 1D , 5A ), which is consistent with previous reports (Yano et al., 2013). Both HD2A and HD2B are essential and functionally redundant for that process since the hd2ahd2b line has a significantly higher DOG1 expression level (25-fold) and stronger dormancy phenotype in comparison to the corresponding single mutants ( Figures 1D , 5A ). Although both, HD2A and HD2B functions, are important for controlling seed dormancy, the HD2B function seems to be more dominant in this process in comparison to HD2A. This is supported by a stronger effect of HD2B knock-down on germination and DOG1 expression than the knock-out of HD2A ( Figures 1C, D , 2 ).

DOG1 is a major genetic factor with a conserved function in controlling seed dormancy (Graeber et al., 2014; Huo et al., 2016). In mature and viable seeds, a higher DOG1 transcript level is associated with stronger dormancy (Bentsink et al., 2006; Nakabayashi et al., 2012). Inter-accession variation of DOG1 expression reflects the dormancy level of seeds of the different accessions. For example, Arabidopsis’s highly dormant accession Cvi has a higher DOG1 expression level than the low-dormant accession Ler (Bentsink et al., 2006). In the highly dormant accession Cvi, the DOG1 expression level is upregulated during seed development and peaked at 16 DAP, and decreased until seed maturation (Nakabayashi et al., 2012). In contrast, in low-dormant accession Col, the DOG1 expression level peaked at 9 DAP and decreased until seed maturation (Zhao et al., 2015), indicating a different regulation mechanism of DOG1 in different dormant accessions. Our result confirmed the earlier decline of DOG1 expression in the low-dormant accession Col ( Figure 5B ). The different DOG1 expression levels might be a consequence of different HD2B expression levels since HD2B directly represses DOG1 by deacetylating the DOG1 coding region ( Figure 7 ). At least, a natural variation of the HD2B expression is described for different Arabidopsis accessions. Arabidopsis highly dormant accession Cvi has a 25-fold lower HD2B expression level in comparison to the low-dormant accession Col (Yano et al., 2013).

We showed that HD2A and HD2B repress DOG1 ( Figure 5A ). The higher HD2B expression during seed maturation contributes to the earlier decline of DOG1 expression in low-dormant accessions. The earlier decline of DOG1 expression, in turn, leads to less DOG1 accumulation in dry seeds and, subsequently, leads to a low-dormant phenotype, such as that of Col. In the hd2ahd2b line, without the repressing function of HD2A and HD2B, DOG1 expression is continuously upregulated reaching the highest expression level in dry seeds ( Figure 5B ). Although DOG1 expression dramatically decreased after imbibition in both, WT and hd2ahd2b, hd2ahd2b seeds still display a significantly higher DOG1 expression than WT ( Figure 5B ). Interestingly, the DOG1 expression level was dramatically increased in fresh dry seeds both in Col ( Figure 5C ) and Cvi accession (Bentsink et al., 2006). The precise mechanism behind that is still unknown. Probably unknown regulators with unique and independent functions from HD2A and HD2B accumulated during that developmental stage or DOG1 mRNA stability is affected.

High DOG1 expression level results in deeper seed dormancy. It was reported that the DOG1 protein level positively correlates with the ABA level in freshly harvested dry and imbibed seeds and negatively correlates with GA biosynthesis during imbibition (Nakabayashi et al., 2012). In this context, enhanced expression of DOG1 in hd2ahd2b ( Figures 5A, B ) would indicate a higher ABA content and lower GA content. Surprisingly, we observed comparable amounts of ABA and GA3 in dry and imbibed seeds of WT and hd2ahd2b ( Figures 3C, D ) concluding that higher DOG1 expression level in hd2ahd2b does not affect ABA and GA3 levels. This is further supported by the unchanged expression of ABA and GA signal transduction-related genes ( Figures 3A, B ). Interestingly, key regulatory genes of ABA and GA biosynthesis and ABA catabolism were upregulated in hd2ahd2b ( Figures 3A, B ), suggesting that HD2A and HD2B may be involved in regulating histone acetylation of these genes.

Reduced expression of HD2A and HD2B resulted in a decrease in total HDA activity ( Figure 6A ), which in turn, led to increased global acetylation of histone H4 and H4K5 ( Figures 6B, C ). However, plant-specific HDAs themselves do most like not possess HDA activity but are rather required for the activity of RPD3-like HDAs. Previous studies (Luo et al., 2012b; Chen et al., 2018) provided evidence that HDAs are acting in multiple protein complexes. HD2A and HD2B seem to be two key subunits of such an HDA complex. Therefore, the loss of HD2A and HD2B function indirectly reduced HDA activity by disturbing the HDA protein complex. DOG1 expression negatively correlated with the expression of HD2A and HD2B ( Figure 5A ). Moreover, the hd2ahd2b line has a significantly higher DOG1 expression level and a stronger seed dormancy phenotype than the corresponding single mutants pointing to an overlapping function of HD2A and HD2B ( Figures 1C, D , 5A, B , S1 ). Loss of DOG1 function in hd2ahd2b genetic background rescued the seed dormancy phenotype ( Figure 5C ), indicating that the upregulation of DOG1 is the reason for the seed dormancy phenotype in hd2ahd2b. HD2B binds to the first exon of DOG1 and deacetylates H4K5 and probably other acetylation marks of H4, too ( Figures 6B, C , 7B, C ). The distance of around 200 - 500 bp downstream of the transcription start (TSS) is typically the regulatory region, where HDAs act. E. g. in S-nitrosoglutathione-treated Arabidopsis seedlings hyperacetylation of H3K9/14 was observed predominantly around 400 bp downstream of TSS (Mengel et al., 2017). Moreover, in hda6 Arabidopsis mutant hyperacetylation of DNA peaked at 200 – 300 bp downstream of TSS (Ageeva-Kieferle et al., 2021).

HD2A and HD2B are interacting with each other ( Figure 8A ) and both plant-specific HDAs may function on DOG1 binding sites. HDAs can be recruited to different DNA binding sites at different developmental stages and in different cell types via different complex partners or DNA binding proteins, such as transcription factors (Liu et al., 2014). Using a BIFC and Co-IP approach, we demonstrated that HD2A and HD2B interact with the transcriptional repressors HSI2 and HSL1 ( Figures 8B, C ), suggesting that during seed development and imbibition, HD2A and HD2B may be recruited by transcriptional repressors HSI2 and HSL1 to DOG1. This results in H4K5 deacetylation of DOG1 and consequently in a decrease in the accessibility of DOG1 for the transcription machinery. It was reported that the selectivity of the HDA activity largely depends on additional modifications of the substrate as well as corepressor binding (Riester et al., 2007; Liu et al., 2014). In yeast, the methyltransferase activity of DOT1 can be specifically activated by H4K16ac sites and is further enhanced by H2B ubiquitination (Valencia-Sánchez et al., 2021). Besides interacting with HD2A and HD2B, HSI2 and HSL1 also recruit CLF and LHP1 for consequent deposition of H3K27me3 marks at DOG1 to inhibit DOG1 expression (Chen et al., 2020). In sum, repression of DOG1 by HSI2 and HSL1 includes a combinatorial regulation via histone acetylation and methylation.

Generally, a fully developed embryo, proper seed storage regents, and well seed coat characteristics (impermeable to water and/or oxygen and low mechanical resistance) are essential for seed dormancy and germination (Focks and Benning, 1998; Wang et al., 2016; Debeaujon et al., 2018). DOG1 plays a central role in regulating seed germination and is also involved in multiple aspects of seed maturation by interfering with ABA signaling components ABI3 and ABI5 (Dekkers et al., 2016).

We demonstrated that HD2A- and HD2B-mediated repression of DOG1 is essential during seed development, maturation, and storage ( Figure 5B ), and loss of HD2A and HD2B function caused multiple defects in seeds ( Figures 9A–H ). Our transcriptomic data provided a general insight into the functions of HD2A and HD2B. GO enrichment analysis demonstrated that these genes are involved in seed maturation, seed dormancy, and seed development process ( Figure 10C ; Supplemental Figure S3 ). It was reported, that DOG1 mediates a conserved coat dormancy mechanism that controls seed germination through the regulation of GA metabolism (Graeber et al., 2014). In this context, it is important to note that the expression of genes involved in “GA biosynthetic processes” and “response to GA” is disturbed in hd2ahd2b ( Figure 10C ). Interestingly, HD2A and HD2B function is also required to control the expression of genes responding to different types of stimuli ( Figure 10C ; Supplemental Figure S3 ). Coordinated responses to external or environmental stimuli are important to cope with environmental changes.

In sum, we showed, that the Arabidopsis plant-specific histone deacetylases HD2A and HD2B have a redundant function and are involved in controlling seed development and germination by coordinating DOG1 expression. Based on our results, we propose a model for the regulatory function of HD2A and HD2B in seed development/germination processes ( Figure 11 ). Acetylation of H4K5 at the 5´-end of the coding region of DOG1 enables its transcription and establishes seed dormancy. During seed maturation and imbibition, HSI2 and HSL1 may recruit HD2A and HD2B to the 5´-end of the coding region of DOG1. Consequently, this region is deacetylated at H4K5 and resulting in the repression of DOG1.