Leaf senescence is the final stage of leaf development. The green leaves gradually turn to yellow, orange, red, and eventually brown and die. This process is accompanied by a series of changes at the cellular, tissue, organ, and organism levels . As a form of programmed cell death (PCD), leaf senescence is primarily an age-dependent process; however, it can also be triggered prematurely by internal and external factors . By integrating environmental and endogenous factors, leaf senescence provides the optimal fitness for plant development .
Leaf senescence is an active rather than passive process to death, and the main functions of leaf senescence are to (a) recycle and re-use the nutrients from senescing leaves into newly developing organs or offspring and (b) enhance the chance of plant survival to adapt to biotic/abiotic stresses [3,4,5]. For grain crops, leaf senescence affects grain yield and quality such as nutrient loss and incomplete filling, etc. . Thus, studying the molecular mechanisms of leaf senescence will not only facilitate the understanding of this fundamental biological process, but may also provide a way to regulate leaf senescence for improving the agricultural traits of crop plants .
To date, many excellent reviews have described the molecular processes involved in leaf senescence in plants [5,6,7,8]. The molecular and genetic understanding of leaf senescence has been mainly gained through the use of the model plant Arabidoposis, which is a Dicot species. Genes controlling leaf senescence are termed as senescence-associated genes (SAGs), and many senescence-associated genes (SAGs) have been identified in plants [9,10]. Our knowledge on the molecular mechanisms underlying leaf senescence in monocots including the major cereals crops such as rice, maize, wheat, barley, and sorghum is still limited. However, with the development of molecular biology and genomics, much inspirational progress has been made in elucidating the molecular mechanisms of leaf senescence in rice. The leaf senescence database currently contains more than 130 SAGs experimentally identified in rice . The objective of this review is to briefly summarize recent progress in this field.
2. Chloroplast Degradation Involved in Leaf Senescence
During leaf senescence, chloroplasts are the first organelles to be dismantled, which can induce the production of reactive oxygen species (ROS) such as hydrogen peroxide (H2O2), superoxide anion radicals (O2−), hydroxyl radicals (OH·), and singlet oxygen (1O2). As signaling triggers, ROS influence the expression of nuclear genes, thereby causing oxidative stress and damage to the cell [12,13,14,15]. The chlorophyll (Chl) degradation pathways involved in leaf senescence have been well established in recent years . Based on the current literature, several SAGs are reported to relate to chlorophyll degradation in rice by using mutations that exhibit a stay-green phenotype in the process of leaf senescence (Table 1). During the degradation of chlorophyll, the first step is the conversion from Chl b to Chl a. NON-YELLOW COLORING 1 (NYC1), a chlorophyll b reductase for catalyzing the degradation chlorophyll b, encodes a chloroplast-localized short-chain dehydrogenase/reductase (SDR) and plays an important role in the degradation of the light-harvesting complex II (LHC II) and the thylakoid membrane (Figure 1) [17,18]. NYC1-LIKE (NOL), a thylakoid membrane location protein, is functionally similar to NYC1. NOL and NYC1 may form a complex to function as a chlorophyll b reductase in rice (Figure 1) . Next, Chl a degradation may start with the de-chelation of Mg2+ by a magnesium-chelating substance which then removes phytol by pheophytinase (PPH) . In rice, NON-YELLOW COLORING 3 (NYC3), which encodes a plastid-localizing α/β hydrolase-fold family protein with an esterase/lipase motif, may function in removing phytol residues from pheophytin a . The STAY GREEN RICE (SGR) gene encodes a senescence-inducible chloroplast stay-green protein 1. The SGR mutant showed chlorophyll retention, stable chlorophyll-protein complexes, and thylakoid membrane structures, but lost its photosynthetic competence during leaf senescence. Further research showed that SGR may be involved in regulating or participating in the activity of pheophorbide a oxygenase (PAO), thereby influencing the degradation of chlorophyll and pigment-protein complexes (Figure 1) . NYC4 (ortholog of Arabidopsis THF1) is also involved in the degradation of chlorophyll–protein complexes during leaf senescence, but its function is distinct from SGR. NYC4 is mainly involved in the degradation of chlorophyll-protein complexes, rather than in the regulation of chlorophyll breakdown . As the downstream of SGR, PAO, and red chlorophyll catabolite reductase (RCCR) are the keys in catalyzing chlorophyll degradation. Knockdown of OsPAO and OsRCCR1 increased the production of ROS, resulting in leaf death and lesion mimic spots (Figure 1) .
The chloroplast degradation mutants above-mentioned all showed a stay-green phenotype; however, Jiao et al.  identified a mutant rapid leaf senescence 1 (rls1), which displayed a rapid leaf senescence during chloroplast degradation. RLS1 encodes an NB-containing protein with an ARM domain at the carboxyl terminus. The NB domain consists of three motifs and is found in many plant disease resistance proteins .
Galactolipids digalactosyl diacylglycerol (DGDG) and monogalactosyl diacylglycerol (MGDG) are the most abundant lipids of thylakoid membranes . At the early stage of leaf senescence, the thylakoid membrane gradually breaks down, and the photosynthetic apparatus disassembles . Osh69, a family of glycosyl hydrolases, encodes alkaline α-galactosidase. The Osh69 protein can cleave the terminal α-galactosidic bond of the galactolipid DGDG . In addition, Osh69 upregulation can be induced by many factors including darkness, hormones, and stress .
3. Phytohormones and Transcription Factors Involved in Rice Leaf Senescence
Phytohormones play vital roles in plant development including leaf senescence (Table 1) .
In rice, the plant hormone methyl jasmonate (MeJA) and its precursor jasmonate (JA) were the first identified senescence promoting substances . CORONATINE INSENSITIVE 1b (OsCOI1b) encodes a homolog of the Arabidopsis jasmonate (JA) receptor COI1. The mutation of OsCOI1b showed methyl jasmonate (MeJA) insensitivity and delayed leaf senescence . By using a metabolite-based genome-wide association study (mGWAS), Fang et al.  identified two major quantitative genes OsPME1 (encoding pectin esterase) and OsTSD2 (encoding pectin methyltransferase) that affected the content of JA. Pectin methyl esterfication is the major source of MeOH. Subsequent investigations using mutants and transgenic lines revealed an MeOH–jasmonates cascade and its epigenetic that regulates leaf senescence . F-box proteins are components of E3 ubiquitin ligase with functions in a wide variety of biological processes . OsFBK12, encoding an F-box protein containing a kelch repeat motif, was involved in 26S proteasome-mediated degradation by interacting with Oryza sativa S-PHASEKINASE-ASSOCIATED PROTEIN1-LIKE PROTEIN (OSK) and targeted the substrate S-ADENOSYL-L-METHIONINE SYNTHETASE1 (SAMS1), triggering changes in ethylene (ETH) levels for the regulation of leaf senescence . ORYZA SATIVA PREMATURE LEAF SENESCENCE (OsPLS1) encoding a vacuolar H+-ATPase subunit A1, plays a negative regulatory role in the onset of rice leaf senescence. The ospls1 mutant showed higher salicylic acid (SA) levels, increased ROS accumulation, and upregulation of WRKY genes . In addition, Yamada et al.  found that strigolactone (SL)-deficient mutants in rice, such as d10, d17, and d27, showed accelerated dark-induced leaf senescence, implying that SL is involved in leaf senescence.
Several senescence-related transcription factors (TFs) are important for regulating leaf senescence (Table 1), for example, the zinc finger transcription factor OsGATA12, whose overexpression causes delayed leaf senescence, the reduction of leaf and tiller number, and improved rice yield. Further study showed that OsGATA12 may be involved in decreased chlorophyll degradation . Overexpression of OsWRKY42 showed an accumulation of ROS and promoted leaf senescence by repressing OsMT1d expression via binding its W-box promoter in rice . The class III homeodomain-leucine zipper (HD-Zip III) gene family plays important roles in plant growth and development . Knockdown of an HD-Zip III member, OsHox33, accelerates leaf senescence in rice . ONAC106, a senescence-associated NACs (NAM/ATAF1/ATAF2/CUC2) transcription factor, negatively regulates leaf senescence .
Many studies have clearly shown that transcription factors and phytohormones interactively regulate the leaf senescence process (Table 1). NACs are plant-specific transcription factors and some NACs have been confirmed to play important roles in regulating leaf senescence [51,52,53,54]. In rice, OsNAP/PS1 encodes a plant-specific NAC transcriptional activator and is induced specifically by abscisic acid (ABA). Overexpression of OsNAP/PS1 significantly promoted premature leaf senescence, whereas knockdown of OsNAP/PS1 produced an obvious delay of leaf senescence . The transcription factor SUBMERGENCE1A (SUB1A), a key regulator of submergence in rice, significantly delays dark-induced senescence by the restriction of MeJA responsiveness and ETH production . A nuclear-localized zinc finger/CCCH transcription factor protein OsDOS (delay of the onset of senescence) was found to take parts of the JA pathway. Overexpression of OsDOS showed delayed leaf senescence, whereas knockdown caused accelerated age-dependent leaf senescence, indicating it was a negative regulator for leaf senescence . In contrast, the rice OsTZF1, which encodes a zinc finger CCCH type family protein, is induced by many factors including ABA, JA, SA, drought, high-salt, and H2O2. Overexpression of OsTZF1 showed delayed seed germination, growth retardation, delayed leaf senescence, improved tolerance to high-salt and drought stresses, and caused opposite phenotypes . OsMYC2, a JA-inducible basic helix-loop-helix transcriptional factor, is a positive regulator of leaf senescence by the direct regulation of some SAGs in rice. Overexpression of OsMYC2 significantly promoted leaf senescence and a reduction in chlorophyll content, and was negatively regulated by OsJAZ8 (a JA ZIM-domain protein), involved in the JA signaling pathway in rice . In addition, a recent study showed that miR319-controlled TCP transcription factors were involved in regulating JA content and leaf senescence .
Aside from the transcription factors mentioned above, based on microarray data, Liu et al. (2016) concluded that the W-box and G-box cis-elements may function as positive regulators affecting rice leaf senescence (Table 1) .
4. Energy Metabolism Pathway Regulated Rice Leaf Senescence
Nicotinamide adenine dinucleotide (NAD) and its derivative nicotinamide adenine dinucleo-tide phosphate (NADP) are important energy metabolite pathways involved in redox reactions in living organisms [57,58]. It was shown that NAD depletion could prevent cell death in vivo to maintain the balance of the internal environment . In Arabidopsis, there are two NADP biosynthetic pathways: de novo and the salvage pathway [60,61]. In the salvage pathway, SIR2, an NAD+-dependent histone deacetylase, plays a crucial role in converting NAD to nicotinamide (Nam) [39,40]. In rice, there are two SIR2 homologous genes, OsSRT1 (OsSIRT701) and OsSRT2 (OsSIRT702) . RNA interference of OsSRT1 results in an increase of histone H3K9 acetylation and a decrease of H3K9 dimethylation, H2O2 accumulation, DNA fragmentation, programmed cell death, and mimicking plant lesions, and its overexpression enhances the tolerance of redox . Recent research indicated that OsSRT1 could regulate carbon metabolic flux through the repression of glycolysis by the deacetylation of both histone and glycolytic glyceraldehyde-3-phosphatedehydrogenase (GAPDH) (Figure 2) .
Downstream of NAD, Nam from nicotinate mononucleotide (NaMN) was catalyzed by two enzymes: nicotinamidase and nicotinate phosphoribosyltransferase (NaPRTase) [59,61]. In rice, a mutation of NaPRTase, LTS1, revealed increased concentrations of nicotinate and nicotinamide as well as decreased NAD content. Further research indicated that the decreased NAD repressed the expression of OsSRTs and would result in a lower deacetylation ability of OsSRTs, hence activating senescence-related genes by increasing the acetylation of histone H3K9, leading to leaf senescence in rice (Figure 2) .
5. Nitrogen Remobilization Involved in Rice Leaf Senescence
Nitrogen remobilization increases nitrogen use efficiency and plays an important role in sustainable agriculture. Nitrogen molecules have a major presence in proteins and nucleic acids, and are transported in the form of amino acids (particularly glutamine and asparagine) from the senescence leaves to new parts . The metabolism of glutamate and γ-aminobutyric acid (GABA) plays an important role in nitrogen circulation . During glutamate metabolism, glutamine synthetase catalyzes ammonia and 2-oxoglutarate into glutamine, whereas glutamate synthase (or glutamine 2-oxoglutarate aminotransferase, GOGAT) catalyzes the reversible conversion of glutamine into glutamate [65,66]. In higher plants, GOGAT has two isoforms: Fd-GOGAT and NADH-GOGAT. Fd-GOGAT is predominantly located in the chloroplasts of photosynthetic tissues, and NADH-GOGAT is present in non-photosynthesizing cells . In rice, the gogat1 mutant exhibited chlorosis under natural conditions and less extent premature leaf senescence under low light conditions. Meanwhile, the gogat1 mutant showed a reduced seed setting rate and increased grain protein and amino acid content. This result showed that OsFd-GOGAT plays an important role in nitrogen remobilization during leaf senescence .
The transferring glutamate to succinate via GABA is called the GABA metabolism or GABA shunt . As a temporary storage of nitrogen, enhanced GABA can inhibit the synthesis of glutamine during senescence . GABA:pyruvate-transaminase catalyzes GABA into succinic semialdehyde (SSA). SSA is then catalyzed into succinate by succinic semialdehyde dehydrogenase (SSADH) and goes into a tricarboxylic acid (TCA) cycle . In rice, Osl2, encoding γ-aminobutyric acid (GABA):pyruvate transaminase, is upregulated and plays a key role in nitrogen metabolism during leaf senescence [42,65].
6. Other Genes Involved in Leaf Senescence
Recent research has shown that cell-wall-related genes are involved in the regulation of leaf senescence. The DWARF AND EARLY-SENESCENCE 1 (DEL1) gene encodes a pectate lyase precursor. Loss of function of DEL1 decreased total pectate lyase (PEL) activity, increased the degree of methylesterified homogalacturonan (HG), and perturbed cell wall composition and structure, resulting in triggering ROS activity, thereby leading to leaf senescence .
UDP-N-acetylglucosamine pyrophosphorylase (UAP) is widely distributed in living organisms . Wang et al.  cloned the SPOTTED LEAF 29 (SPL29) gene, which encodes UAP1 in rice. The spl29 mutant displayed many changes involved in chloroplast degradation, chlorophyll loss and photosystem II decline, enhanced resistance to bacterial blight inoculation, increased malondialdehyde content and ROS, upregulated SAGs and defence response genes, downregulated photosynthesis-related genes, etc. .
Actin filament plays an important role in many endomembrane processes such as vacuole formation, endocytosis of plasma membrane (PL), and vesicle transport from the Golgi complex, etc. [68,69,70,71,72]. The ARP2/3 complex as a key regulator of actin filament nucleation can be inactive by itself and active by the SCAR/wave complex in plants . The SCAR/WAVE complex is highly conserved, and deficiency in the SCAR/WAVE complex in plants often leads to morphological changes . In rice, EARLY SENESCENCE 1 (ES1) encodes a SCAR-LIKE PROTEIN2, which plays an important role in leaf senescence. The es1 mutant shows a short and irregular arrangement of actin filaments. The changes to the actin filaments increase the water loss of leaves, thereby leading to leaf senescence .
The SWEET family plays important roles in plant growth and development. In rice, overexpression of OsSWEET5, a novel sugar transporter family, caused growth retardation and precocious senescence at the seedling stage .
Glycine decarboxylase complex (GDC) is a multi-protein complex, which plays a major role in the photorespiration of plants . Under ambient CO2, knockdown of OsGDCH caused leaf senescence due to chlorophyll loss, protein degradation, chloroplast breakdown, and autophagy, as well as ROS accumulation .
Genomic imprinting is an epigenetic phenomenon in mammals that causes parental-specific, monoallelic expression of a subset of autosomal genes. The unique expression patterns of imprinted genes are controlled by allele-specific DNA methylation of the cis-regulatory sequences, called the imprinting control regions (ICRs). Because allelic DNA methylation of ICRs is acquired during gametogenesis, ICRs are also called germline differentially methylated regions (gDMRs) (Ferguson-Smith, 2011; Kelsey and Feil, 2013; Tomizawa and Sasaki, 2012). Recent genome-wide profiling has, however, revealed that the number of oocyte- or sperm-specific methylated genomic regions is far more than that of the known ICRs (Kobayashi et al., 2012; Smallwood et al., 2011). Therefore, germline methylation might not be restricted to ICRs, and both the ICRs and the other gDMRs could be methylated by a common mechanism without being strictly discriminated in the germ line (Kelsey and Feil, 2013).
During preimplantation development, although most gDMRs lose their gamete-derived methylation via epigenetic reprogramming activity, allelic methylation of ICRs is faithfully maintained (Kobayashi et al., 2012; Smallwood et al., 2011). Presumably, the methylation maintenance mechanism against genome-wide demethylation activity might operate at restricted genomic loci and selected alleles in preimplantation embryos. We and others have reported that Stella (also known as DPPA3) (Nakamura et al., 2007), KAP1 (also known as TRIM28) (Messerschmidt et al., 2012), and NuRD complex components (Ma et al., 2010; Reese et al., 2007), in addition to the maintenance methyltransferase DNMT1 (Hirasawa et al., 2008), help to maintain ICR methylation in preimplantation embryos. Because these factors have no sequence specificity for DNA binding or intrinsic DNA binding ability themselves, other sequence-specific DNA binding proteins must be required for their recruitment to specific target sites. The Krüppel-associated box (KRAB)-containing zinc-finger protein ZFP57, which was found to interact with KAP1, is a dominant candidate for such a protein (Li et al., 2008; Messerschmidt et al., 2012; Quenneville et al., 2011). However, because the depletion of ZFP57 did not always affect the methylation of ICRs (Li et al., 2008), other factors and their target cis elements are apparently engaged in the maintenance mechanism.
The H19 ICR in the mouse Igf2/H19 locus (Fig. 1A) is DNA-methylated by the DNMT3A-DNMT3L complex in prospermatogonia, the status of which is maintained on the paternal allele following fertilization (Kaneda et al., 2004; Tremblay et al., 1997), and it is thus classified as a gDMR. Whereas indispensable roles for CTCF (Matsuzaki et al., 2010; Schoenherr et al., 2003) and Sox-Oct binding motifs (Sakaguchi et al., 2013; Zimmerman et al., 2013) in maintaining maternal H19 ICR hypomethylation during postimplantation periods are well established, little is known about the underlying mechanisms that maintain paternal H19 ICR hypermethylation during preimplantation periods.
We previously tested the activity of the H19 ICR in yeast artificial chromosome (YAC) transgenic mice (TgM), in which an H19 ICR fragment (2.9 kb) was inserted into a YAC bearing the (nonimprinted) human β-globin locus (150 kb, Fig. 1B) to minimize position effects of transgene insertion sites (Tanimoto et al., 2005). In somatic cells, the H19 ICR fragment was preferentially methylated when paternally inherited, demonstrating that the 2.9-kb sequence contained sufficient information to recapitulate imprinted methylation. Surprisingly, however, the transgenic H19 ICR was not methylated in the testes. In addition, randomly integrated H19 ICR fragments in the mouse genome were hypermethylated in the paternal allele after fertilization, irrespective of their variable methylation levels in the testes of multiple TgM lines (Matsuzaki et al., 2009). It was therefore presumed that the H19 ICR was marked by an epigenetic modification other than DNA methylation in the germ line, and that paternal allele-specific methylation was acquired after fertilization by referring to this hypothetical mark. Hence, our results and those of others (Gebert et al., 2010; Park et al., 2004) suggested that two distinct methylation mechanisms operate at the endogenous H19 ICR: one in the germ line, which is under the control of its surrounding sequences, and the other during the postfertilization period, which is governed by a hypothetical epigenetic mark preset within the H19 ICR during gametogenesis. We speculated that the latter activity might be actively involved in the region-specific maintenance of allelic methylation at the endogenous H19 ICR in preimplantation embryos.
In this study, we show that the paternal-allele-specific methylation of the transgenic H19 ICR commences soon after fertilization in YAC-TgM, and that maternally supplied DNMT3A and DNMT3L are required for this process. By partially obstructing germline methylation of the endogenous H19 ICR, we discovered that postfertilization methylation activity also exists at the endogenous H19 ICR. Furthermore, in YAC-TgM, we substantially narrowed the responsible sequences for postfertilization methylation acquisition in the transgenic H19 ICR. Finally, by deleting the responsible sequences from the endogenous locus, we noted a partial loss of methylation in the paternally inherited H19 ICR after fertilization, diminished Igf2 expression, and embryonic growth retardation in the offspring that paternally inherited the mutation. These results demonstrate that the postfertilization methylation imprinting activity of the H19 ICR is essential for maintaining its imprinted methylation status once established during gametogenesis.
Methylation acquisition at the transgenic H19 ICR in early embryos
The H19 ICR fragment inserted into the β-globin YAC transgene (Fig. 1B) exhibited preferential DNA methylation in the somatic cells of offspring after paternal transmission, whereas it was not methylated in sperm (Fig. S1A-C) (Tanimoto et al., 2005). As a first step in elucidating the mechanism of the allele-specific methylation of the transgenic H19 ICR, we examined the timing of its acquisition in mouse early embryos. Bisulfite sequencing of the transgenic H19 ICR (region I′ including the CTCF site 1, Fig. 1B) revealed that the paternally inherited ICR was moderately and heavily methylated in one- and two-cell stage embryos, respectively (Fig. 1C,D, Fig. S1D), the levels of which were substantially higher than in the maternally inherited alleles. While the DNA methylation level in region I′ in two-cell embryos was already high (Fig. 1D, Fig. S1D) and indistinguishable from that in blastocyst-stage embryos (Fig. S1E) (Matsuzaki et al., 2010), DNA methylation around CTCF binding site 4 of the paternally inherited H19 ICR was low in blastocysts (region II, Fig. S1F), suggesting that DNA methylation acquisition directionally extends from a region near CTCF site 1. These results suggested that the paternally inherited transgenic H19 ICR is recognized by the DNA methylation machinery soon after fertilization and becomes progressively methylated during embryonic development.
DNA methylation status of the transgenic H19 ICR in early embryos. (A) Structure of the mouse Igf2/H19 locus. Mouse Igf2 and H19 (open boxes) are ∼90 kb apart, and the expression of both genes depends on the shared 3′ enhancer (gray box). The H19 ICR, located approximately at −4 to −2 kb relative to the transcription start site of H19 is contained within a 2.9-kb SacI (Sa)-BamHI (B) fragment. Dots (1-4) indicate the position of CTCF binding sites. G; BglII site. (B) Structure of the ICR/β-globin YAC transgene. The 150-kb human β-globin locus YAC carries the LCR (gray box) and the β-like globin genes (open boxes). The 2.9-kb H19 ICR fragment (inverted orientation) was introduced between the LCR and the ε-globin gene (Tanimoto et al., 2005). A gray bar below the map indicates sequences (region I′) analyzed by bisulfite sequencing. (C,D) Methylation status of the transgenic H19 ICR in embryos. One- (C) or two-cell (D) embryos that inherited the ICR/β-globin YAC transgene (line 1048) either paternally (Pat.) or maternally (Mat.) were embedded in agarose beads (19-43 embryos per bead in C, 13-37 embryos per bead in D) and treated with sodium bisulfite. The beads were separately and directly used to amplify the region I′ of the transgenic H19 ICR by nested PCR. PCR products were individually subcloned and sequenced. The results from single beads are presented together in a cluster. Each horizontal row represents a single DNA template molecule. The numbers on the right of each row indicate number of times the pattern was observed in the sequencing. Methylated and unmethylated CpG motifs are shown as filled and open circles, respectively. (E,F) A role of de novo DNA methyltransferases in the postfertilization methylation of the paternally inherited transgenic H19 ICR. Two-cell (E) or blastocyst (F)-stage embryos were obtained from wild-type (WT), [Dnmt3l −/−], [Dnmt3a2lox/2lox, Zp3-Cre], or [Dnmt3b2lox/2lox, Zp3-Cre] females crossed with the ICR/β-globin male TgM carrying WT Dnmts (5-24 embryos in E, 1-9 embryos in F). The methylation status of the paternally inherited transgenic H19 ICR (region I′) was analyzed by bisulfite sequencing as described previously. (G) E8.5 embryos were obtained from WT or [Dnmt3l−/−] females crossed with ICR/β-globin male TgM. Genomic DNA was extracted from each embryo and treated with sodium bisulfite, and region I′ of the transgenic H19 ICR was amplified by nested PCR. PCR products were subcloned and sequenced. The results from single embryos are presented together in a cluster. Above each panel in E-G are genotypes of mothers.
A role for DNMT3s in the postfertilization methylation of the transgenic H19 ICR
Methylation acquisition in the endogenous H19 ICR occurs in fetal prospermatogonia via the actions of DNMT3A and DNMT3L (Kaneda et al., 2004). However, the activity and targets, if any, of the DNMT3 family in early embryos remain obscure. We thus examined which DNMTs were involved in methylation acquisition in the transgenic H19 ICR. Because the gene products present in early embryos soon after fertilization are mostly derived from oocytes, we assessed the roles of DNMTs on postfertilization methylation of the transgenic H19 ICR after maternal disruption. To test the function of Dnmt3l, Dnmt3l-null (−/−) (Hata et al., 2002) females were mated with ICR/β-globin YAC transgenic (Dnmt3l wild-type) (Fig. 1B) (Tanimoto et al., 2005) males. Because Dnmt3a−/− or Dnmt3b−/− mice are not viable (Okano et al., 1999), we used Cre-loxP recombination to specifically eliminate these genes in growing oocytes via the zona pellucida glycoprotein 3 (Zp3) promoter-Cre transgene (de Vries et al., 2000; Dodge et al., 2005; Kaneda et al., 2004). After confirming that Dnmt3 gene products were depleted in both oocytes and early embryos by quantitative reverse transcription-polymerase chain reaction (RT-qPCR) (Fig. S2), we analyzed the methylation status of the transgenic H19 ICR fragment. In Dnmt3l-deficient two-cell embryos, the paternally inherited transgenic H19 ICR was hypomethylated (Fig. 1E). Depletion of maternally provided Dnmt3a gene product also caused hypomethylation of the transgenic H19 ICR, whereas the loss of the Dnmt3b gene product did not affect its methylation (Fig. 1E). These results demonstrated that the postfertilization methylation acquisition of the paternally inherited transgenic H19 ICR required both DNMT3A and DNMT3L, which were maternally provided to early embryos.
We next examined whether zygotic expression of Dnmts (Fig. S2) (Guenatri et al., 2013; Hu et al., 2008) would compensate for a loss of maternally provided DNMT3L in transgenic H19 ICR methylation during embryogenesis. The paternally inherited transgenic H19 ICR remained unmethylated in blastocyst-stage embryos derived from Dnmt3l−/− mothers (Fig. 1F). The unmethylated state of the transgenic H19 ICR did not change even at embryonic day (E) 8.5, despite the fact that allele-nonspecific methylation, probably elicited by postimplantation de novo DNA methylation activity, was observed outside of the DMR (Fig. 1G) (Matsuzaki et al., 2010). These results demonstrated that the paternally inherited transgenic H19 ICR must be recognized by the DNA methylation machinery, including DNMT3A and DNMT3L, in early embryos to acquire imprinted methylation.
Evaluation of the postfertilization methylation activity at the endogenous H19 ICR
Although the transgenic H19 ICR possesses intrinsic activity to acquire paternal allele-specific methylation in early embryos, it is unclear whether this activity also exists at the endogenous locus. Because the endogenous H19 ICR is fully methylated in sperm, the postfertilization methylation activity at the endogenous locus, if present, is normally difficult to reveal. Our previous results (Matsuzaki et al., 2009) and those of others (Gebert et al., 2010; Park et al., 2004; Puget et al., 2015) suggested that the gametic methylation of the H19 ICR was governed by signals from surrounding sequences, i.e. those located outside the 2.9-kb H19 ICR region. Therefore, by interfering with the transmission of these hypothetical signals and subsequent methylation during spermatogenesis, we sought to verify the postfertilization methylation imprinting activity at the endogenous locus. To this end, we inserted tandemly arrayed chicken HS4 core sequences, (cHS4c)2,on both sides of the endogenous H19 ICR (Fig. 2A), expecting that this would block a presumptive signal to direct DNA methylation of the H19 ICR in the germ line, as the cHS4 itself was unmethylated in both germ and somatic cells when it was substituted for the endogenous H19 ICR (Szabo et al., 2002). Importantly, during the postfertilization period, the same manipulation does not prevent methylation imprinting activity in the context of YAC-TgM (Okamura et al., 2013a). Embryonic stem cells (ESCs) were modified by homologous recombination, and accurate targeting events were confirmed by Southern blotting (Fig. S3A,B). Following the establishment of two knock-in mouse lines (Fig. S3B), the neor selectable marker was excised by mating them with Cre-expressing TgM (Fig. S3C).
DNA methylation status of the insulated H19 ICR in sperm and blastocysts. (A) Map of the wild-type and mutant H19 loci. Tandem cHS4 core fragments (gray rectangles, I for insulator) were inserted at both sides of the H19 ICR [at BglII (G) sites: mutant allele] (see also Fig. S3). Three regions of the mutant allele (III, IV, and V; gray bars beneath the map) were analyzed by bisulfite sequencing in B and C. B, BamHI; Sa, SacI sites; dots 1-4, CTCF binding sites. (B) Genomic DNA extracted from sperm of mutant mice (lines 61 and 97) was digested with XbaI and treated with sodium bisulfite. Three regions of the mutant allele were amplified by PCR, subcloned, and sequenced. The overall percentage of methylated CpGs in region IV or those in the ICR portions of region III or V are indicated next to each panel. (C) Blastocyst-stage embryos that inherited the mutant allele paternally were embedded in agarose beads (6-12 embryos per bead) and treated with sodium bisulfite. The beads were used for amplifying three regions of the mutant allele by PCR, and the resulting fragments were individually subcloned and sequenced.
Germline methylation of the endogenous H19 ICR is inhibited by flanking insulator sequences
We examined the methylation status of the insulated H19 ICR allele in sperm by bisulfite sequencing. The (cHS4c)2 sequences on both sides of the H19 ICR were hypomethylated (Fig. 2B). In addition, the H19 ICR region containing CTCF sites 3 and 4 was also methylated at very low levels. Furthermore, the region around the CTCF sites 1 and 2 was significantly less methylated (Fig. 2B) in comparison to the fully methylated sequences in the wild-type allele (Fig. S3D) (P<0.0001, Mann–Whitney U-test; http://quma.cdb.riken.jp/). Southern blotting using methylation-sensitive restriction enzymes confirmed these results (Fig. S4A,B). These results indicated that the flanking (cHS4c)2 fragments at the endogenous H19 ICR inhibited its methylation acquisition during spermatogenesis. It was reported that USF1 binding to cHS4 sequences induces histone H3/H4 acetylation and H3K4 methylation, thereby interfering with the spread of repressive histone modifications (such as H3K9 methylation) and heterochromatin formation (West et al., 2004). We therefore infer that intrusion of the repressive chromatin state into the H19 ICR from outside might be a prerequisite for its gametic methylation and that flanking cHS4c might block this process. Alternative hypotheses might be that VEZF1-bound cHS4c sequences or simply its CpG island-like nature somehow interfered with the de novo DNA methylation of the neighboring H19 ICR only during spermatogenesis (Dickson et al., 2010; Kobayashi et al., 2012; Smallwood et al., 2011). It is also possible that cHS4c sequences interfered with transcription read-through at the H19 ICR, which has been suggested to induce its methylation in male germ cells (Henckel et al., 2012).
Allele-specific postfertilization DNA methylation at the insulated H19 ICR
We next examined the methylation status of the insulated H19 ICR after fertilization. In blastocyst-stage embryos (Fig. 2C), as well as in the somatic cells of neonates (Fig. S4A,C-E), the paternally inherited insulated H19 ICR exhibited higher methylation levels than those observed in sperm (Fig. 2B). The markedly elevated level of methylation in (cHS4c)2 sequences on the paternal allele (Fig. S4E) was probably a secondary consequence of imprinted methylation of the adjacent H19 ICR as the same sequences were unmethylated after maternal transmission. These results demonstrated that the endogenous H19 ICR was methylated after fertilization in a paternal allele-specific manner, and that the methylation acquisition commenced during the preimplantation period.
In the YAC-TgM, transgenic H19 ICR methylation in early embryos was dependent on maternally provided DNMT3L and DNMT3A (Fig. 1E,F). We therefore examined whether the same DNA methyltransferases were operating at the endogenous insulated H19 ICR. When Dnmt3l −/− female mice were bred, the paternally inherited insulated H19 ICR region (region IV in Fig. 3A) was less methylated in blastocyst (Fig. 3B) than in control embryos (WT), indicating that DNMT3L is involved in the methylation of the endogenous H19 locus. In addition, the effect of maternal DNMT3L depletion on the postfertilization methylation at the endogenous insulated H19 ICR was observed in as early as 2-cell embryos (Fig. 3C), the same timing when postfertilization methylation acquisition took place in the YAC TgM (Fig. 1D).
Function of DNMT3L in postfertilization methylation of the paternally inherited insulated H19 ICR. Blastocyst (lines 61 and 97) and 2-cell (line 61) stage embryos were obtained from wild-type (WT) or [Dnmt3l−/−] females crossed with male mice carrying mutant (insulated) H19 ICR allele. DNA methylation status of region IV (shown in A, map legend as per Fig. 2A) of the paternally inherited mutant allele was analyzed by bisulfite sequencing, as described above, for blastocyst- (B) and 2-cell stage embryos (C). Three to 10 (blastocyst) or 13 to 30 (2-cell) embryos per agarose bead were used. Above each panel in B and C are genotypes of mothers.
Taken together, these results suggest that postfertilization paternal allele-specific methylation activity also exists in the endogenous H19 ICR, which is probably governed by a shared mechanism with the transgenic H19 ICR.
Defining the H19 ICR DNA sequences essential for allele-specific postfertilization DNA methylation in YAC-TgM
To further elucidate the mechanisms underlying the postfertilization methylation of the H19 ICR, we sought to more precisely define its responsive sequences. To this end, we employed a YAC-TgM system, in which the methylation activity is clearly detectable after fertilization because of its unmethylated state in sperm. We previously demonstrated that the 1.7-kb ‘ICR21’ fragment covering CTCF sites 1 and 2 (Fig. 4A) was sufficient to recapitulate paternal allele-specific DNA methylation after fertilization in YAC-TgM (Okamura et al., 2013b). We thus generated a series of 5′-truncated H19 ICR fragments: the ICR4321S fragment, in which the 5′-end of the 2.9-kb H19 ICR fragment (766 bp) was deleted but all four CTCF sites remained; and the ICR432 fragment, which is 173 bp shorter than the ICR4321S fragment and lacks CTCF site 1 (Fig. 4A). To reduce the time required to obtain mouse lines carrying intact single-copy YAC transgenes, these two fragments were individually floxed using hetero-specific loxP sequences, combined to employ a co-placement strategy (Tanimoto et al., 2008), and introduced 3′ to the locus control region (LCR) in the human β-globin YAC (Fig. S5A). YAC-TgM were generated by pronuclear injection and intact, single-copy transgene carriers were identified (Fig. S5A,B). Parental YAC-TgM lines (Numbers 443 and 429) were crossed with Cre-TgM to initiate in vivo Cre-loxP recombination, which generated daughter lines carrying either the ICR4321S or ICR432 transgene at the identical chromosomal integration site (Fig. 4B, Fig. S5C,D).