ROS1-Dependent DNA Demethylation Is Required for ABA-Inducible NIC3 Expression^,

ROS1-mediated DNA demethylation is a prerequisite to the transcriptional activation of NIC3 in response to ABA.

Abstract

DNA methylation plays an important role in diverse developmental processes in many eukaryotes, including the response to environmental stress. Abscisic acid (ABA) is a plant hormone that is up-regulated under stress. The involvement of DNA methylation in the ABA response has been reported but is poorly understood. DNA demethylation is a reverse process of DNA methylation and often induces structural changes of chromatin leading to transcriptional activation. In Arabidopsis (Arabidopsis thaliana), active DNA demethylation depends on the activity of REPRESSOR OF SILENCING 1 (ROS1), which directly excises 5-methylcytosine from DNA. Here we showed that ros1 mutants were hypersensitive to ABA during early seedling development and root elongation. Expression levels of some ABA-inducible genes were decreased in ros1 mutants, and more than 60% of their proximal regions became hypermethylated, indicating that a subset of ABA-inducible genes are under the regulation of ROS1-dependent DNA demethylation. Notable among them is NICOTINAMIDASE 3 (NIC3) that encodes an enzyme that converts nicotinamide to nicotinic acid in the NAD⁺ salvage pathway. Many enzymes in this pathway are known to be involved in stress responses. The nic3 mutants display hypersensitivity to ABA, whereas overexpression of NIC3 restores normal ABA responses. Our data suggest that NIC3 is responsive to ABA but requires ROS1-mediated DNA demethylation at the promoter as a prerequisite to transcriptional activation. These findings suggest that ROS1-induced active DNA demethylation maintains the active state of NIC3 transcription in response to ABA.

As sessile organisms, plants are constantly exposed to a variety of harsh environmental conditions such as an excess or a deficit of light, water, and temperature. Plants perceive and cope with such abiotic stresses through signal transduction pathways accompanied by changes in the expression of stress-responsive genes (Yamaguchi-Shinozaki and Shinozaki, 2006; Mirouze and Paszkowski, 2011; Kinoshita and Seki, 2014; Zhu, 2016). Abscisic acid (ABA) is a key phytohormone crucial for growth and development including seed dormancy, stomatal movement, and environmental stress responses. Upon exposure to drought and high salinity, ABA is accumulated in plant cells, followed by transcriptional activation of stress-responsive genes through ABA signaling networks (Nambara and Marion-Poll, 2005; Finkelstein, 2013; Mittler and Blumwald, 2015; Yoshida et al., 2015).

Epigenetic regulation involving DNA methylation, histone modification, and noncoding RNAs triggers mitotically or meiotically heritable changes in gene expression through modulation of chromatin structure without DNA sequence changes. In response to environmental stress, epigenetic mechanisms contribute to changes in stress-responsive gene expression patterns, which allows plants to perceive and memorize stress through mitotic or meiotic transmission of epigenetic states (Chinnusamy and Zhu, 2009; Mirouze and Paszkowski, 2011; Heard and Martienssen, 2014; Wibowo et al., 2016; Ganguly et al., 2017). Several studies report that histone modification, a primary epigenetic mechanism including acetylation, methylation, phosphorylation, and ubiquitination on the N-terminal tails of histone proteins, is crucial for ABA, drought, and salt stress responses (Chinnusamy and Zhu, 2009; Kinoshita and Seki, 2014). It was also reported that Arabidopsis (Arabidopsis thaliana) mutants defective in histone deacetylase activity were hypersensitive to ABA and salt, whereas the plants overexpressing histone deacetylase exhibited ABA insensitivity and increased tolerance to drought and salt stresses (Sridha and Wu, 2006; Chen et al., 2010; Cui et al., 2013).

DNA methylation is a stable but reversible epigenetic mark responsible for diverse biological processes such as cell differentiation, gene imprinting, genome stability, and X chromosome inactivation (Law and Jacobsen, 2010; Zhang and Zhu, 2012; Smith and Meissner, 2013; Wu and Zhang, 2014). Although DNA methylation in mammals primarily occurs in the CG context, DNA methylation in plants is found at cytosine bases in all sequence contexts: symmetric CG and CHG (where H is A, C, or T) and asymmetric CHH (Law and Jacobsen, 2010). In plants, de novo methylation is established by DOMAINS REARRANGED METHYLTRANSFERASE 2 through a small interfering RNA–mediated pathway, which is known as RNA-directed DNA methylation (Matzke and Mosher, 2014). DNA methylation is maintained by three DNA methyltransferases acting on different sequence contexts: DNA METHYLTRANSFERASE 1, CHROMOMETHYLASE 3, and DOMAINS REARRANGED METHYLTRANSFERASE 2 at CG, CHG, and CHH sequences, respectively (Zhang and Zhu, 2012; Pikaard and Mittelsten Scheid, 2014). In plants, DNA methylation can be actively removed by the plant-specific DNA glycosylases/lyases, which directly excise 5-methylcytosine (5mC), resulting in unmethylated cytosine through the base excision repair pathway (Zhu, 2009; Wu and Zhang, 2014). Four members of the 5mC DNA glycosylase family are found in Arabidopsis - REPRESSOR OF SILENCING 1 (ROS1), DEMETER (DME), DEMETER-LIKE 2 (DML2), and DML3 (Choi et al., 2002; Penterman et al., 2007). All these DME/ROS1 family proteins directly excise 5mC from DNA, while cleaving the sugar-phosphate backbone by successive β- and δ-elimination reactions in vitro (Agius et al., 2006; Gehring et al., 2006; Morales-Ruiz et al., 2006; Penterman et al., 2007; Ortega-Galisteo et al., 2008; Lee et al., 2014). However, they play distinct roles in plant development. For instance, DME is preferentially expressed in the central cell of the female gametophyte and responsible for seed development and endosperm gene imprinting (Choi et al., 2002; Gehring et al., 2006; Hsieh et al., 2011), whereas ROS1, DML2, and DML3 are expressed in the vegetative tissue with their biological functions poorly understood (Penterman et al., 2007). ROS1 was initially identified in a genetic mutant screen for transcriptional silencing of the ABA- and stress-responsive Responsive to Desiccation 29A::LUCIFERASE (RD29A::LUC) transgene and the endogenous RD29A gene via promoter hypermethylation (Gong et al., 2002). It is postulated that ROS1, along with DML2 and DML3, is necessary for preventing excessive DNA methylation at genomic regions, and its loss-of-function mutation leads to transcriptional silencing of transgenes as well as the endogenous genes (Gong et al., 2002; Penterman et al., 2007; Lister et al., 2008; Qian et al., 2012; Stroud et al., 2013).

Recent studies provided evidence that DNA methylation plays essential roles in stress responses. Genome-wide analyses revealed that biotic stress-induced dynamic DNA methylation changes of repetitive sequences or transposable elements were associated with differentially expressed genes, which were also up-regulated in met1 and drm1 drm2 cmt3 (ddc) mutants, explaining enhanced resistance to the bacterial pathogen in met1 and ddc mutants (Dowen et al., 2012). Additionally, in response to abiotic stress, widespread DNA methylation alterations at transposable elements took place in rice after nearby gene expression changes with partial mitotic heritability (Secco et al., 2015). In particular, ROS1-mediated active demethylation is required for expression of EPIDERMAL PATTERNING FACTOR2 (EPF2), which regulates stomatal development in response to ABA-mediated drought stress via the regulation of transpiration rate (Yamamuro et al., 2014). Hyperosmotic stress induces DNA methylation changes at specific genomic regions associated with transmission of salt tolerance to the progeny through the female germline (Wibowo et al., 2016).

In this study, we investigated the relationship between DNA methylation and ABA-mediated stress responses. Here we show that ros1 mutants were hypersensitive to ABA for early seedling establishment and root elongation. To understand the effect of the ros1 mutation in ABA responsiveness accompanied with changes in global DNA methylation patterns, the transcriptome and DNA methylome were comparatively analyzed between wild-type and ros1 mutants. Nearly 80% of differentially expressed genes (DEGs) were down-regulated in ros1-4 mutants. Notably, more than 60% of the proximal regions of the ABA-inducible DEGs were found to be hypermethylated in ros1 mutants. This suggests that down-regulation of a subset of ABA-inducible genes in ros1 mutants is associated with excessive DNA methylation, leading to ABA hypersensitivity. Remarkably, NICOTINAMIDASE 3 (NIC3), which encodes an enzyme that catalyzes the deamination of nicotinamide (Nam), is linked to the ABA responsiveness. In ros1 mutants, ABA inducibility of NIC3 expression was severely compromised, associated with DNA hypermethylation of transposable elements (TEs) proximal to the promoter region compared to the wild type. The nic3 mutants exhibited hypersensitivity to ABA; and moreover, ectopic expression of NIC3 in ros1 mutants restored normal responses to ABA in early seedling development and root growth. Together, our findings suggest that decreased expression of NIC3 associated with DNA hypermethylation can contribute to ABA hypersensitivity of ros1 mutants.

RESULTS

The ros1 Mutants Exhibit Hypersensitivity to ABA

ABA sensitivity screening on seed germination, early seedling establishment, and root growth has been widely used to identify genes involved in ABA metabolism and signaling (Finkelstein, 2013; Yoshida et al., 2014; Zhu, 2016). To investigate the effect of DNA methylation on transcriptional regulation of ABA-responsive genes, genetic screens were performed on Arabidopsis T-DNA insertion mutants defective in DNA methylation and demethylation with respect to ABA responsiveness. We found that two ros1 T-DNA insertion lines ros1-3 and ros1-4 were hypersensitive to ABA for early seedling establishment and root growth (Fig. 1; Supplemental Fig. S1). In the absence of ABA, no significant difference was found between the wild type and ros1 mutants. On the other hand, in the presence of ABA, the wild type successfully established early seedlings and displayed only slightly reduced primary root growth (Finkelstein, 2013; Yoshida et al., 2014), whereas ros1 mutants failed to develop early seedlings and exhibited a significant decrease in root growth in an ABA concentration-dependent manner (Fig. 1). These observations indicate that ros1 mutants are hypersensitive to ABA.

Figure 1.

ABA hypersensitivity of ros1 mutants. A, Cotyledon greening phenotypes of the wild type (WT) and ros1 at 12 d after sowing (DAS) on MS medium without or with ABA. B, Quantitative analysis of green cotyledon emergence at different ABA concentrations every 4 d. Data points are the percentage of green cotyledons against total seeds on MS medium (n > 80). C, Root growth phenotypes at 19 DAS on MS medium without or with ABA. Red lines indicate average root lengths for each genotype. D, Quantitative analysis of root growth at 19 DAS with different concentrations of ABA. Error bars indicate ± sd (n = 4). Asterisks indicate significant difference from the wild type (Student’s t test with Bonferroni-Holm correction; * P < 0.05). Scale bar = 10 mm.

The ros1 Mutation Alters Gene Expression Patterns upon ABA Treatment

ROS1 is responsible for repressing transcriptional silencing of some endogenous genes including RD29A, EPF2, and stress-responsive genes via the removal of DNA methylation marks (Gong et al., 2002; Le et al., 2014; Yamamuro et al., 2014). Given the putative function of ROS1 in regulating gene expression patterns, we hypothesized that some genes important for ABA responses may be dysregulated in ros1, which leads to ABA hypersensitive phenotypes in the mutants. To test this hypothesis, RNA sequencing (RNA-seq) analysis was performed on wild type and ros1 mutants with the criteria of a twofold change cutoff and a 5% false discovery rate. We identified a total of 75 and 89 DEGs between the wild type and ros1-4 in the absence and presence of ABA, respectively (Fig. 2, A and B). Nearly 80% of DEGs were down-regulated in ros1-4, suggesting that the ros1 mutation may lead to transcriptional silencing of target genes. The ros1-3 mutation also resulted in a total of 113 and 140 DEGs in the absence and presence of ABA, among which ∼80% and 60% of them were down-regulated, respectively, in ros1-3 (Supplemental Fig. S2). EPF2 was down-regulated in ros1-4, consistent with the previous report (Yamamuro et al., 2014), but not in ros1-3. We thus focused on analyzing the transcriptome data in ros1-4. We further verified the RNA-seq data by performing reverse transcription-quantitative PCR (RT-qPCR) for the representative DEGs (Supplemental Fig. S3).

Figure 2.

Transcriptome changes caused by the ros1-4 mutation in the absence or presence of ABA. A, Scatter plot of gene expression levels in the wild type (WT) and ros1-4 as determined by RNA-seq. The mRNA expression level was calculated as log₂-scaled fragments per kilobase of exon per million mapped fragments (FPKM). Red, green, and black dots indicate up-regulated genes, down-regulated genes, and non-DEGs, respectively. B, The Venn diagram of down-regulated and up-regulated genes in ros1-4. C, Hierarchical clustering of all 116 DEGs in ros1-4 categorized into six clusters.

To cluster the genes with distinct expression patterns in response to ABA in the ros1 background, we performed hierarchical clustering analysis for 116 DEGs between the wild type and ros1-4 (Fig. 2C; Supplemental Table S1). All DEGs were grouped into six clusters according to their expression characteristics. Cluster 1 (n = 12) and Cluster 2 (n = 9) include the genes up-regulated in ros1-4 in the ABA-dependent and -independent manners, respectively. Cluster 3 (n = 19) represents the genes whose expression is ABA-inducible in the wild type but not in ros1-4 (Table 1; Supplemental Fig. S4). Cluster 4 (n = 38) includes the genes down-regulated by the ros1 mutation independent of ABA treatment. Cluster 5 (n = 31) includes the genes whose expression is repressed by ABA in the wild type but not in ros1. EPF2 is also found in cluster 5. Cluster 6 (n = 7) includes the genes down-regulated by ABA both in the wild type and ros1-4. We rationalized that the candidate genes responsible for ABA hypersensitivity in ros1 might result from their loss of ABA inducibility, and therefore, genes in cluster 3 were subjected to further in-depth analysis.

Table 1. List of a subset of genes in cluster 3.

(–), no data; FC, log₂-scaled fold change; FDR, false discovery rate; FPKM, fragments per kilobase of transcript per million mapped reads.

Gene IDa	Symbola	Wild Type				ros1-4				Descriptiona
		FPKM		FC	FDR	FPKM		FC	FDR
		−ABA	+ABA			−ABA	+ABA
AT1G58270b	ZW9	43.91	156.78	1.84	1.85E-55	12.36	33.48	1.44	3.31.E-24	TRAF-like family
AT2G02120b	LCR70	42.51	242.56	2.51	2.15E-28	33.13	82.88	1.32	4.32.E-08	Scorpion toxin-like knottin superfamily
AT5G22860b	–	7.40	12.24	0.73	4.61E-06	2.98	4.81	0.69	4.92.E-03	Ser carboxypeptidase S28 family
AT2G18193b	–	6.38	10.25	0.68	3.64E-02	3.27	3.53	0.11	1	P-loop containing nucleoside triphosphate hydrolases superfamily
AT5G23220b	NIC3	4.54	29.06	2.67	1.00E-31	5.72	10.17	0.83	3.38.E-03	Nicotinamidase 3
AT5G25120b	CYP71B11	2.56	5.11	1.00	4.14E-04	0.96	1.48	0.62	0.30	Cytochrome P450, family 71, subfamily B, polypeptide 11
AT1G43590b	–	1.40	2.30	0.72	0.62	0	0	0.00	1	Transposable element gene
AT2G01580	–	1.40	3.00	1.10	2.26E-02	0.63	1.01	0.67	0.70	Unknown
AT2G06002	–	1.10	2.69	1.28	0.11	0	0	0.00	1	Other RNA
AT1G52990b	–	0.35	1.72	2.25	4.71E-04	0.24	0.42	0.79	1	Thioredoxin family
AT5G08250b	–	0.30	1.00	1.72	4.40E-03	0.06	0.16	1.29	0.69	Cytochrome P450 superfamily
AT2G26750	–	0.17	1.08	2.59	1.40E-03	0.10	0.20	1.01	0.92	α/β-Hydrolases superfamily
AT5G28520	–	0.13	49.27	8.44	3.05E-128	0.07	13.63	7.34	1.81.E-59	Man-binding lectin superfamily

^aInformation is adopted from TAIR10 (http://arabidopsis.org/).

^bGene proximal to the ros1-hyper DMRs within 2 kb.

The ros1 Mutation Contributes to DNA Hypermethylation

To investigate changes in genome-wide DNA methylation patterns in ros1, whole-genome bisulfite sequencing analysis was carried out. We identified differentially methylated cytosines (DMCs) and differentially methylated regions (DMRs) in ros1-3 and ros1-4 compared with the wild type (Supplemental Tables S2–S4). In accordance with the previous data (Huang et al., 2013; Tang et al., 2016), both in ros1-3 and ros1-4, the numbers of hypermethylated DMCs (hyper-DMCs) and hyper-DMRs were significantly higher than those of hypomethylated (hypo-) DMCs and DMRs, respectively (Supplemental Figs. S5 and S6; Supplemental Tables S2–S4). The relative prevalence of 171,550 common hyper-DMCs in ros1-3 and ros1-4 (ros1 hyper-DMCs) was examined for each sequence context, revealing that CG hypermethylation predominantly occurred in ros1 mutants (Fig. 3A). Nearly 70% of the ros1 hyper-DMCs were found in the CG context, whereas ∼15% each of them were in the CHG and CHH contexts, respectively (Fig. 3A). We also identified 4,044 common hyper-DMRs in ros1-3 and ros1-4 (ros1 hyper-DMRs; Fig. 3B; Supplemental Fig. S5). The ros1 hyper-DMRs in each sequence context substantially overlapped each other, and the number of the CHH DMRs was less than that of CG and CHG DMRs (Fig. 3B). Consistent with the observation that 70% of the CG DMRs overlapped with CHG DMRs, CG DMRs exhibited higher CHG DNA methylation levels, and CHG DMRs also exhibited higher CG methylation levels (Fig. 3, B and C). Hence, these results suggest that DNA methylation levels were considerably increased in ros1 compared with the wild type, a substantial portion of which occurred in both CG and CHG contexts.

Figure 3.

Genome-wide DNA methylation changes in ros1 mutants. A, Fraction of the ros1 hyper-DMCs as determined by whole genome BS-seq. B, Overlap of the ros1 hyper-DMRs in each sequence context. C, Heat maps of DNA methylation levels for the ros1 hyper-DMRs. Each horizontal line indicates the 5mC level of single DMR. D, Distribution of the ros1 DMCs and DMRs along the Arabidopsis chromosomes. The red lines and black bars are, respectively, the numbers of DMCs and DMRs in a bin (500 kb). The positive and negative numbers represent the counts of hyper- and hypo-DMCs/DMRs, respectively. Gray bars indicate pericentromeric regions. WT, wild type.

In comparison with the previous data (Qian et al., 2012; Stroud et al., 2013; Wang et al., 2013), DNA methylation levels of the ros1 hyper-DMRs were found to be similar to those of ros1 or ros1 dml2 dml3 (rdd) triple mutants, distinct from other mutants defective in genetic components involved in DNA methylation (Supplemental Fig. S7; Supplemental Table S5). Distribution of ros1 DMCs and DMRs across the genome revealed that the ros1 hyper-DMCs and DMRs occurred at high density on chromosome arms but at low density in pericentromeric regions (Fig. 3D), consistent with the previous studies (Penterman et al., 2007; Qian et al., 2012). Thus, these observations indicated that ROS1 may demethylate discrete loci in euchromatin regions in the Arabidopsis genome.

Down-Regulation of ABA-Inducible Genes in ros1 Is Associated with an Increase in DNA Methylation Levels

To clarify the relationship between the ros1–down-regulated genes and the ros1-hyper DMRs, we counted the number of genes proximal to the ros1-hyper DMRs within 2 kb. The ros1-hyper DMRs were proximal to 5,692 genes (18.3% of Arabidopsis coding genes), among which 52 genes were down-regulated in ros1. Remarkably, 63.2% (12 of 19) of the genes in cluster 3 were associated with the ros1-hyper DMRs (Table 1). Because of the close relationship between ros1-hyper DMRs and transcriptional repression of the genes in cluster 3, we further analyzed the expression levels and the DNA methylation status of the genes in this group (Fig. 4, A and B). Four representative genes in cluster 3 were found to be ABA-inducible in the wild type, but the induction was abolished in ros1 (Fig. 4A). In particular, in the absence of ABA, AT2G02120 and NIC3 were not down-regulated in ros1, but in the presence of ABA, they were significantly down-regulated in ros1 (Fig. 4A). In addition, the ros1-hyper DMRs were present in the upstream region of AT2G02120 and NIC3 (Fig. 4B) and in the downstream region of AT1G58270 and AT2G43670, indicating that DNA methylation levels of their proximal regions were increased in ros1 compared with the wild type. These results indicate that lack of ABA-inducibility of the genes in cluster 3 highly correlates with DNA hypermethylation, leading to ABA hypersensitive phenotypes in ros1.

Figure 4.

Association of decreased expression levels with DNA hypermethylation in the genes in cluster 3. A, The log₂-scaled fold change in gene expression of the representative genes in cluster 3. Error bars indicate ± sd (n = 3). Hashes and asterisks indicate significant differences from the wild type (WT) in the absence and presence of ABA, respectively (Student’s t test with Bonferroni-Holm correction; *P < 0.05). B, Genome browser views of the DNA methylation levels at the representative genes in cluster 3 and their proximal genomic regions. The left side is the upstream region of the locus. Positive and negative bars indicate 5mC levels of single cytosine on the Watson (+1) and Crick (-1) strands, respectively. Guide bar = 500 bp.

DNA Hypermethylation at the NIC3 Promoter Is Accompanied with Decreased NIC3 Expression

Our RNA-seq and whole genome BS-seq analysis revealed that a subset of ABA-inducible genes were down-regulated in ros1 associated with excessive DNA methylation. Most notable among the candidate genes in cluster 3 was NIC3, which encodes a catalytic enzyme that converts Nam to nicotinic acid in the NAD⁺ salvage pathway (Fig. 4, A and B). Nicotinamidase Pnc1 in yeast is probably the most famous for its role as an activator of Silent information regulator 2 (Sir2), a NAD⁺-dependent histone deacetylase critical for transcriptional silencing at the rDNA locus under calorie restriction, heat stress, and salt stress conditions (Anderson et al., 2003; Gallo et al., 2004). The Arabidopsis genome contains three nicotinamidase genes NIC1, NIC2, and NIC3. The nic1 and nic2 mutants were reported for their hypersensitivity to ABA, salt stress, and Nam, suggesting that nicotinamidase play crucial roles in ABA-mediated osmotic stress responses (Hunt et al., 2007; Wang and Pichersky, 2007). Despite high sequence similarity to NIC2, functions of NIC3 have not been elucidated.

The RT-qPCR analysis showed that the NIC3 expression level was similar in both wild type and ros1 mutants without ABA treatment (Fig. 5A). In contrast, upon ABA treatment, NIC3 expression was significantly induced in wild type but to a much lesser degree in ros1 (Fig. 5A). Unlike NIC3, expression levels of both NIC1 and NIC2 were similar in the wild type and ros1, while displaying no responsiveness to ABA (Supplemental Fig. S8). Tissue-specific expression analysis revealed that NIC3 expression was highest in the roots, followed by imbibed seeds, flower buds, and siliques (Fig. 5B). NIC3 expression was relatively low in rosette leaves and cauline leaves (Fig. 5B).

Figure 5.

Suppression of ABA-inducible NIC3 expression by DNA hypermethylation at the upstream region of the NIC3 promoter in ros1 mutants. A, Relative expression levels of NIC3 in the wild type (WT) and ros1 in the absence and presence of ABA as determined by RT-qPCR. Error bars indicate ± sd (n = 3). B, Relative expression levels of NIC3 in 24-h imbibed seeds (IS), roots (RT), rosette leaves (RL), cauline leaves (CL), flower buds (FB), and young siliques (SL) of 5-week-old Arabidopsis plants by RT-qPCR analysis. Error bars indicate ± sd (n = 3). C, Schematic diagram of the NIC3 locus. Lines indicate intergenic regions, whereas white and black boxes, respectively, indicate untranslated regions and coding regions. Relative locations of the promoter TEs (gray boxes) and DMRs (red bars) are presented. D, DNA methylation levels at the DMR I and DMR II in the wild type and ros1 as determined by local BS-seq.

DNA hypermethylation of the upstream NIC3 promoter was then investigated for its functional implications in suppression of ABA-inducible NIC3 expression in ros1. The upstream region of NIC3 promoter harbors four TEs, two of which contained the ros1-hyper DMRs (Fig. 5C). The local BS-seq analysis revealed that the first DMR (DMR I) detected at the RC/Helitron family of TE (AT5TE28305) exhibited an increase in DNA methylation levels for all sequence contexts (Supplemental Fig. S9). The second DMR (DMR II) detected at the DNA/MuDR family of TE (AT5TE28295) displayed an increase of only CG methylation (Fig. 5D; Supplemental Fig. S9). These findings strongly suggest that DNA hypermethylation of TEs at the upstream of NIC3 promoter is responsible for suppression of its expression and ABA hypersensitivity in ros1 mutants.

Ectopic Expression of NIC3 Mitigates ABA Hypersensitivity of ros1

To determine whether NIC3 expression is crucial for ABA responses, two loss-of-function mutant lines nic3-1 and nic3-2 were examined for ABA sensitivity in early seedling development and root growth. For early seedling establishment, nic3 mutants were hypersensitive to ABA compared to the wild type, despite their phenotype being weaker than ros1 (Fig. 6, A and B). Moreover, root growth of nic3 mutants was inhibited in a manner similar to ros1 upon ABA treatment (Fig. 6, C and D). Consistently, nic3 mutants exhibited increased sensitivity to ABA with respect to early seedling establishment and primary root elongation, suggesting that the decrease of NIC3 expression in ros1 may contribute to ABA hypersensitivity. To verify the function of NIC3 in converting Nam into nicotinic acid, we also investigated the effects of nic3 and ros1 mutations on early seedling growth and development upon Nam treatment. In accordance with the previous reports on developmental defects of nic1 and nic2 mutants (Hunt et al., 2007; Wang and Pichersky, 2007), nic3 mutants also failed to develop green cotyledons by Nam treatment compared with the wild type and ros1, probably because of the failure of Nam turnover (Supplemental Fig. S10). However, primary root growth of nic3, ros1, and wild type was similarly inhibited by Nam treatment, indicating that the accumulation of Nam affected root elongation independently of NIC3.

Figure 6.

ABA hypersensitivity of nic3 mutants. A, Cotyledon greening phenotypes of the wild type (WT), ros1, and nic3 at 12 d after sowing (DAS) on MS medium without or with ABA. B, Quantitative analysis of green cotyledon emergence every 4 d. Data points are the percentage of green cotyledons against total seeds on MS medium (n > 70). C, Root growth phenotypes at 19 DAS on MS medium without or with ABA. Red lines indicate average root lengths for each genotype. D, Quantitative analysis of root growth at 19 DAS. Error bars indicate ± sd (n = 4). Asterisks indicate significant difference from the wild type (Student’s t test with Bonferroni-Holm correction; *P < 0.05). Scale bar = 10 mm.

We next investigated whether NIC3 overexpression could restore the ABA sensitivity in ros1. Transgenic plants were generated in the ros1-4 background by overexpressing NIC3 under the constitutive CaMV 35S promoter (NIC3ox-1, and -2; Fig. 7, A and B). Upon ABA treatment, two independent T2 NIC3ox lines restored normal early seedling establishment and root elongation in a similar manner to the wild type (Fig. 7, C–F). This finding indicates that ectopic expression of NIC3 was able to alleviate the ABA hypersensitive phenotypes of ros1 for early seedling development and root growth, suggesting that the transcriptional active state of NIC3 is achieved by ROS1-mediated DNA demethylation in ABA responses.

Figure 7.

Restored early seedling development and root growth by ectopic expression of NIC3 in ros1 upon ABA treatment. A, Schematic diagram of constitutive NIC3 expression by the CaMV 35S promoter. B, Relative expression of NIC3 in the wild type (WT), ros1-4, and two NIC3ox lines in the ros1-4 background (NIC3ox-1,2). Error bars indicate ± sd (n = 3). C, Cotyledon greening phenotypes at 15 or 17 d after sowing (DAS) on MS medium without or with ABA, respectively. D, Quantitative analysis of green cotyledon emergence every 4 d. Data points are the percentage of green cotyledons against total seeds on MS medium (n > 70). E, Root growth phenotypes at 18 or 23 DAS on MS medium without or with ABA, respectively. Red lines indicate average root lengths for each genotype. F, Quantitative analysis of root growth at 19 DAS. Error bars indicate ± sd (n = 4). Asterisks indicate significant difference from the wild type (Student’s t test with Bonferroni-Holm correction; *P < 0.05). Scale bar = 10 mm.

DISCUSSION

DNA methylation generally represses the expression of proximal genes associated with TEs (Slotkin and Martienssen, 2007; Lisch, 2013; Diez et al., 2014). ROS1, a member of the DNA demethylase family in Arabidopsis, has been proposed to be responsible for locus-specific demethylation rather than indiscriminate genome-wide demethylation (Zhu et al., 2007; Tang et al., 2016). Recent studies reported that ROS1-mediated DNA demethylation is necessary for the expression of EPF2 and stress-responsive genes for stomatal development and Fusarium oxysporum resistance, respectively, by preventing the spread of DNA methylation from the upstream TEs (Le et al., 2014; Yamamuro et al., 2014). This study also demonstrated that ROS1-initiated DNA demethylation at the TEs in the NIC3 promoter region is crucial for its ABA inducible expression, suggesting the role of NAD⁺ biosynthesis and dynamic nicotinamide level changes in ABA responses.

ROS1 is required for the expression of the ABA- and stress-responsive RD29A::LUC transgene and the endogenous RD29A gene, suggesting the role of ROS1 in ABA-mediated stress responses (Gong et al., 2002). In the absence of ABA, ros1 mutants did not exhibit obvious developmental defects, consistent with the previous reports (Fig. 1; Gong et al., 2002). In the presence of ABA, however, ros1 mutants were hypersensitive to ABA for early seedling establishment and root growth in an ABA-concentration dependent manner, indicating that ROS1-mediated DNA demethylation is required for ABA responses (Fig. 1). Although the detailed mechanism by which ROS1 responds to environmental changes such as ABA up-regulation and osmotic stress remains elusive, the ROS1-dependent DNA demethylation pathway may transmit environmental cues to the genome, thereby providing prompt responsiveness to the plant cell.

Consistent with the proposed role of ROS1 in transcriptional silencing associated with DNA demethylation (Gong et al., 2002; Penterman et al., 2007; Lister et al., 2008; Qian et al., 2012; Stroud et al., 2013), the transcriptome and methylome analysis revealed that a significant number of DEGs were down-regulated in ros1 mutants compared with the wild type, accompanied with hyper-DMRs (Figs. 2 and 3). In support of our hypothesis that down-regulation of ABA-inducible genes may result in ABA-hypersensitive phenotypes of ros1, more than 60% of the genes in cluster 3 were found to be located proximal to the hyper DMRs in ros1 mutants. This indicates a close relationship between the reduced level of gene expression and DNA hypermethylation (Table 1). It is noteworthy that the nic3 mutants were hypersensitive to ABA with regard to early seedling establishment and root growth (Fig. 6). Remarkably, DNA methylation levels of the TEs in the upstream region of NIC3 promoter (AT5TE28305 and AT5TE28295) were increased in ros1 mutants, thereby leading to the decrease in NIC3 expression (Fig. 5). Hence, these observations suggest that ROS1 prevents spreading of DNA methylation from nearby TEs to the NIC3 promoter, ensuring the expression of NIC3 in response to ABA. However, considering the occurrence of numerous DMRs in ros1 mutants, we cannot rule out the possibility that ROS1 is required for the activation of the factor(s) involved in ABA-dependent induction of NIC3, rather than its direct activation. Moreover, a total of 1,489 genes were up-regulated upon ABA treatment in the wild type, and most of them (1,203 of 1,489) still showed normal ABA inducibility in ros1-4 (Gene Expression Omnibus [GEO] accession number GSE65016 at National Center for Biotechnology Information [NCBI]). Approximately 84% (1,009 of 1,203) of these genes whose transcription levels were increased byABA treatment both in the wild type and ros1-4 were not proximal to ros1-hyper DMRs. Representative ABA-inducible genes such as RD29B and MYB DOMAIN PROTEIN 41 also exhibited normal ABA induction in ros1 mutants without hyper-DMRs in their proximal regions, suggesting that ABA inducibility is also controlled in a ROS1-independent manner (Supplemental Fig. S11).

Nicotinamidase is a catalytic enzyme that converts Nam into nicotinic acid in the NAD⁺ salvage pathway. Many enzymes in the NAD⁺ salvage pathway such as poly(ADP-ribose) polymerase and nicotinate/nicotinamide mononucleotide adenyltransferase are known to be involved in stress responses (De Block et al., 2005; Vanderauwera et al., 2007; Hashida et al., 2010). Our findings suggest that dynamic regulation of Nam and NAD⁺ might be necessary for ABA-dependent developmental processes in Arabidopsis. Further investigation on the relationship between the NAD⁺ salvage pathway and the ABA response will provide the intriguing but currently undisclosed connection between plant stress hormone signaling and NAD⁺ metabolism, because the cellular level of NAD⁺/NADH is regarded as a metabolic rheostat to measure the energy production activity of living cells.

Pnc1, a homolog of NIC3 in yeast, is an activator of a NAD⁺-dependent histone deacetylase Sir2, and functions to lower the cellular Nam levels resulting in extended lifespan in response to calorie restriction, heat, and hyperosmotic stresses (Anderson et al., 2003; Gallo et al., 2004). The stress-responsive transcription factors Multicopy suppressor of sucrose non-fermenting 1 mutation 2 and Multicopy suppressor of sucrose non-fermenting 4 in yeasts are responsible for PNC1 expression in response to hyperosmotic shocks (Ghislain et al., 2002; Medvedik et al., 2007). It remains to be addressed whether NIC3 plays a similar regulatory role to SIR2 or SIR2-like factors for ABA responses in plants. In contrast with yeasts that do not use DNA methylation for epigenetic regulation of gene expression, plants may engage ROS1-mediated DNA demethylation for ABA responses, allowing adaptation to environmental changes under a rather dynamic control.

MATERIALS AND METHODS

Plant Materials

Arabidopsis (Arabidopsis thaliana) ecotype Columbia (Col-0) was used as the wild type control in this study. The ros1-3 seeds were provided (Penterman et al., 2007). The homozygous T-DNA insertion lines including ros1-4 (SALK_045303), nic3-1 (SALK_034040), and nic3-2 (SALK_107343) were obtained from Arabidopsis Biological Resource Center. Seeds were sterilized and stratified at 4°C for 3 d in the dark. They were sown on Murashige and Skoog (MS) medium, then grown in a growth chamber at 22°C under 16 h of fluorescent light at 30 ± 10 μmol m⁻² s⁻¹.

Cotyledon Greening and Root Elongation Assays

To determine the sensitivity to ABA or Nam, the MS medium was supplemented with the indicated concentrations of ABA or Nam. Seedlings with developed green cotyledons were counted every 4 d. For root elongation assay, 4-d-old seedlings grown on the MS medium were transferred to MS medium containing different concentrations of ABA or Nam. Seedlings were allowed to grow vertically for additional days, and then the primary root length was measured.

Plant RNA Extraction, RNA Sequencing (RNA-seq), and Analysis

For each of the three biological replicates, five plants were pooled to obtain a single RNA sample. Total RNA was isolated with the RNeasy plant RNA extraction kit (Qiagen) and treated with RQ1 RNase-free DNase (Promega) to eliminate any genomic DNA contaminants. RNA-seq libraries were constructed according to the previous report (Zhong et al., 2011). RNA-seq was performed on the Illumina HiSeq 2500 sequencing system.

An average of ∼4.7 m paired-end reads (2×101 bp) was obtained from each library. The subsequent quality trimming and artificial sequence elimination steps were performed manually. Total RNA-Seq reads were mapped to the Arabidopsis TAIR10 genome using TopHat software (v2.0.8; Trapnell et al., 2009). To analyze gene expression, the raw count data were collected using the htseq-count command of HTSeq (v0.6.1p; Anders et al., 2015). The DEGs were identified by the edgeR package (v3.5.25; Robinson et al., 2010) with the criteria of a twofold change cutoff and 0.05 false discovery rate. The hierarchical clustering was performed by the pheatmap package (Kolde, 2015), and a heat map of gene expression profiles were generated by cluster analysis.

RT-qPCR

The complementary DNA synthesis was conducted using the QuantiTech Reverse Transcription kit (Qiagen). Quantitative RT-PCR was performed using the Roter-Gene Q (Qiagen) with SYBR green Q-master mix (Genet Bio). Each experiment consisted of three biological replicates and was repeated twice. UBIQUITIN10 was used as an internal control for the calculation of relative transcript level of target genes. Primers used for RT-qPCR are listed in Supplemental Table S6.

Whole Genome Bisulfite Sequencing and Analysis

A total of 5 μg of Arabidopsis genomic DNA was used to generate BS-seq libraries. The library construction and sequencing steps were performed with Illumina HiSeq 2000. An average of ∼43.4 m paired-end reads (2×101 bp) was obtained from each library. Low-quality sequences (q < 20) were trimmed manually, and the trimmed reads were mapped to the TAIR 10 genome using bwa-meth (v0.09; Pedersen et al., 2014) and Bis-SNP (v. 0.82.2; Liu et al., 2012), with the parameters “–map-q 60 -T BisulfiteGenotyper -C CG,1 -C CHG,1 -C CHH,1 -out_modes EMIT_ALL_CYTOSINES”. Only cytosine sites with 4× coverage were used for subsequent analysis. DMCs and DMRs were identified as described previously (Huang et al., 2013). DMRs were finally identified based on the regions with a length ≥ 100 bp, ≥ 5 DMCs, and the mean methylation difference ≥ 0.3 for CG, ≥ 0.15 for CHG, or ≥ 0.1 for CHH. Published BS-seq data from various mutants defective in genetic components of DNA methylation and demethylation were retrieved from NCBI GEO (Supplemental Table S5).

Local Bisulfite Sequencing

A total of 1 μg of Arabidopsis genomic DNA was subjected to bisulfite treatment and used for the subsequent sequencing analysis. The bisulfite treatment and DNA purification steps were performed with the Epitech Bisulfite kit (Qiagen). Each region-of-interest was amplified from the bisulfite-treated DNA, and the individual PCR products were sequenced in more than quintuplicate. The results were analyzed and visualized with aid of CyMATE (Hetzl et al., 2007). The used oligomer sets are presented in Supplemental Table S6.

Plant Expression Vector Construction and Plant Transformation

For the 35S::NIC3 construct, the coding sequence of NIC3 was amplified from complementary DNA and cloned into the pGHX vector under the control of cauliflower mosaic virus (CaMV) 35S promoter (Fujita et al., 2012). The verified construct was introduced into the Agrobacterium tumefaciens strain GV3101. Transgenic Arabidopsis plants in the ros1-4 background were generated by Agrobacterium-mediated floral dip method. T1 plants were selected on MS medium containing 30 mg L⁻¹ hygromycin with 25 mg L⁻¹ cefotaxime. The verified T2 lines were used for further analyses. The used oligomer sets are presented in Supplemental Table S6.

Accession Numbers

The RNA-seq and DNA methylome data generated in the study were deposited at the NCBI GEO under accession numbers GSE65016 and GSE64569, respectively.

SUPPLEMENTAL DATA

The following supplemental materials are available.

• Supplemental Figure S1. Characterization of ros1 mutants.

• Supplemental Figure S2. Transcriptome changes caused by the ros1-3 mutation in the absence or presence of ABA.

• Supplemental Figure S3. Correlation of the RNA-seq data with RT-qPCR analysis.

• Supplemental Figure S4. Loss of ABA inducibility of the genes in cluster 3 in ros1-4 as determined by RNA-seq.

• Supplemental Figure S5. Overlap of DNA methylation changes caused by the ros1-3 or ros1-4 mutation.

• Supplemental Figure S6. DNA methylation changes in ros1 on chromosome 1.

• Supplemental Figure S7. The 5mC levels of the ros1 hyper-DMRs in mutants defective in DNA methylation and demethylation.

• Supplemental Figure S8. Expression patterns of Arabidopsis NIC genes as determined by RNA-seq.

• Supplemental Figure S9. DNA methylation levels of NIC3.

• Supplemental Figure S10. Characterization of nic3 mutants.

• Supplemental Figure S11. Normal ABA induction of a subset of genes in ros1-4 without hyper-DMRs.

• Supplemental Table S1. List of cluster genes.

• Supplemental Table S2. Estimated differentially methylated regions (DMRs) from ros1-3 vs. WT.

• Supplemental Table S3. Estimated differentially methylated regions (DMRs) from ros1-4 vs. WT.

• Supplemental Table S4. Estimated commonly differentially methylated regions (DMRs) from ros1 vs. WT.

• Supplemental Table S5. Accession numbers of BS-seq data from other studies.

• Supplemental Table S6. List of primers.

Dive Curated Terms

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

• DML2 Gramene: AT3G10010

• DML2 Araport: AT3G10010

• DML3 Gramene: AT4G34060

• DML3 Araport: AT4G34060

• NIC3 Gramene: AT5G23220

• NIC3 Araport: AT5G23220

• NIC2 Gramene: AT5G23230

• NIC2 Araport: AT5G23230

• AT2G02120 Gramene: AT2G02120

• AT2G02120 Araport: AT2G02120

• NIC1 Gramene: AT2G22570

• NIC1 Araport: AT2G22570

• MET1 Gramene: AT5G49160

• MET1 Araport: AT5G49160

• AT1G58270 Gramene: AT1G58270

• AT1G58270 Araport: AT1G58270

• AT2G43670 Gramene: AT2G43670

• AT2G43670 Araport: AT2G43670

• ros1 Gramene: At2g36490

• ros1 Araport: At2g36490

• ros1-4 Gramene: At2g36490

• ros1-4 Araport: At2g36490