Molecular dissection of complex agronomic traits of rice: a team effort by Chinese scientists in recent years

Rice is a staple food for more than half of the worldwide population and is also a model species for biological studies on monocotyledons. Through a team effort, Chinese scientists have made rapid and important progresses in rice biology in recent years. Here, we briefly review these advances, emphasizing on the regulatory mechanisms of the complex agronomic traits that affect rice yield and grain quality. Progresses in rice genome biology and genome evolution have also been summarized.


domestication, rice
Rice is a staple food worldwide and more than half of the global population uses rice as the main food source. In addition, rice is an excellent model species for studies on monocotyledonous plants due to its small genome size, the completed genome sequence and vast genetic resources. During the early years of the 21st century, Chinese scientists launched on their Long March to understanding the complex agronomic traits of rice, initiated by the completion of the draft genome sequence of the indica variety 93-11 and a fine sequence analysis of chromosome 4 of the japonica variety Nipponbare [1,2]. These progresses have made crucial contributions to the international efforts in sequencing the rice genome. Shortly afterward, the identification of MONOCULM1 (MOC1), a key gene controlling rice shoot development [3], marked an important step for Chinese rice biologists in their astonished efforts and achievements during the past 10 years. Since then, researchers have made rapid and impressive progresses in almost every aspect of rice biology by employing the genetics-based strategy in combination with multidisciplinary approaches. Some of these advances have been comprehensively reviewed [4][5][6][7]. Here, we briefly summarize the progresses in rice biology made by Chinese scientists in recent years, focusing on the studies of several important agronomic traits and genome biology.

MOLECULAR BASIS OF SHOOT ARCHITECTURE FORMATION
In rice, the shoot architecture is mainly determined by the number of shoots (known as tillers or culms in rice) per plant, tiller angle and plant height. Shoot branching is a common feature of higher plants, and this process is usually referred to as tillering in grasses. In rice, a tiller is initiated on the unelongated basal internode as an axillary bud at the leaf axil, followed by outgrowth at the later stage independently of the main culm. Because the panicle is formed at the tip of each tiller, the proper control of the tiller number is therefore a key determinant of grain production in rice [8,9]. The molecular mechanism of controlling tillering in rice and other crops remained largely unknown until the identification and characterization of the rice MOC1 gene. The loss-of-function mutations in MOC1 results in a single main culm phenotype, whereas the overexpression of MOC1 leads to the enhanced tillering capacity, thus demonstrating that MOC1 is a master control of tillering in rice [3]. MOC1 encodes a GRAS family transcription factor that is structurally and functionally conserved in both monocotyledons and dicotyledons. A comparative genomic study reveals that the MOC1 locus is highly conserved in both the gene colinearity and the structure among all the examined 14 Oryza species, including 10 diploids and 4 allotetraploids [10]. Moreover, mutations in the LAX PANICLE2 (LAX2) gene causes a similar phenotype as moc1. LAX2, encoding a novel nuclear protein, genetically interacts with MOC1, and a lax2 moc1 double mutant shows an enhanced mutant phenotype in the initiation of axillary buds, suggesting that LAX2 and MOC1 function in independent pathways or in the same protein complex in regulating tillering [11].
Although the critical role of MOC1 in the control of tillering has been well appreciated, two key questions remain unanswered: how the MOC1 activity is regulated and how does MOC1 activate its downstream components? Several recent studies have uncovered important clues on these questions. In two independent studies, three allelic mutants tillering and dwarf 1/tiller enhancer (tad1/te) have dramatically increased tiller number, a phenotype opposite to that of moc1. TAD1/TE encodes a co-activator of the anaphase-promoting complex (APC/C), a multisubunit E3 ubiquitin ligase that promotes the proteasomal degradation of the MOC1 protein in a cell cycle-dependent manner [12,13]. As a master switch of tillering, MOC1 regulates the expression of two downstream transcription factor genes, OSH1 and OsTB1, which are involved in the regulation of meristem initiation and maintenance and axillary bud outgrowth, respectively [3]. In line with this observation, OsTB1 was recently shown to antagonize the activity of OsMADS57, a negative regulator of tillering [14]. OsMADS57 directly targets DWARF14 (D14; also known as D88 and HTD2 for HIGH TILLERING AND DWARF 2) to repress its expression; D14/D88/HTD2 negatively regulates tillering by modulating the signal-ing of the phytohormone strigolactone [15,16]. The transcriptional repression activity of OsMADS57 on D14/D88/HTD2, in turn, is relieved by physically interacting with OsTB1. Notably, the expression of OsMADS57 is also negatively regulated by miR444a [14]. Therefore, MOC1-regulated tillering is likely attributed to, at least partly, modulating strigolactone signaling (Fig. 1).
Tillering is a complex trait regulated by multiple pathways or genes, among which phytohormone signaling plays a critical role. In addition to the classic plant hormones auxin and cytokinin, the newly identified phytohormone strigolactone, derived from carotenoids in the chloroplast, negatively regulates shoot branching or tillering in both monocotyledons and dicotyledons. In rice, strigolactone inhibits axillary bud growth, thereby negatively regulating tillering. Mutations in genes involved in strigolactone biosynthesis and signaling cause the increased tiller number and dwarf phenotype, and these mutants in rice are mainly designated as dwarf (d) mutants. A few DWARF genes have been partially characterized, of which D27 encodes a novel iron-containing β-carotene isomerase that catalyzes all-trans-β-carotene to 9-cis-β-carotene in the strigolactone biosynthetic pathway [17,18]. The d27 phenotype is correlated with the increased polar auxin transport, implying a possible crosstalk between these two phytohormones [17].
In the long-lasting efforts to search for an 'ideal' rice architecture to enhance grain yield, a semidominant quantitative trait locus, IDEAL PLANT ARCHITECTURE1/WEALTHY FARMER'S PAN-ICLE (IPA1/WFP), was characterized. IPA1 encodes a transcription factor OsSPL14, and the expression of OsSPL14 is regulated by OsmiR156. The ipa1 allele carries a mutation at the target site of OsmiR156, resulting in partial deregulation by Os-miR156 and consequently a higher level of IPA1 transcripts. The ipa1 allele renders the formation of an ideal rice architecture featured with the reduced tiller number, increased lodging resistance and enhanced grain yield. In a field test, the grain yield of the variety carrying the ipa1 allele increased more than 10% over its parental line carrying wildtype IPA1 allele [19] (Fig. 1). The ST-12 rice line carrying WFP, an allele of IPA1, was characterized by an independent study. ST-12 possibly contains epigenetic mutations in the IPA1/WFP promoter, thus causing a higher expression level of Os-SPL14 [20]. In a genome-wide ChIP-seq analysis of the binding sites of IPA1, a large number of target genes were identified in shoot apices (1067) and young panicles (2185), of which 581 genes were identified in both tissues. A gene expression profiling analysis by RNA-seq identified more than 10% REVIEW Zuo and Li 255 differential expressed genes associated with IPA1binding targets. Of those, OsTB1 was characterized as a direct target gene of IPA1 that positively regulates OsTB1 to suppress tillering [21] (Fig. 1). The characterization of IPA1/WFP and its targets not only defines ideal rice architecture, but also provides a practical means to enhance rice production.
In addition to the tiller number, the tiller angle is also an important agronomic trait to determine ideal architecture and, eventually, the grain yield. In rice, LAZY1, a novel grass-specific protein, is a major regulator of the tiller angle. LAZY1 is involved in the regulation of gravitropism through negatively regulating polar auxin transport in shoots, and the lazy1 mutation causes an extremely tillerspreading phenotype [22]. The LOOSE PLANT ARCHITECTURE1 (LPA1) gene encodes a plantspecific transcription repressor and acts independently of LAZY1 to regulate the tiller angle and leaf angle by controlling the adaxial growth of the tiller node and lamina joint [23].
Ancestral wild rice adapts the prostrate or tillerspreading growth that is phenotypically similar to that of lazy1, whereas modern rice cultivars become erect growth. The transition from the prostrate growth to the erect growth was a key event in rice domestication. In wild rice, the prostrate growth architecture is determined by a putative zinc-finger transcription factor PROSTRATE GROWTH 1 (PROG1), which shares a similar expression pattern as LAZY1. Intriguingly, in all the examined 182 varieties of cultivated rice, an identical loss-of-function mutation was detected, and the transformation of a wild-type PORG1 transgene into a cultivated rice variety renders the erect growth into the prostrate growth [24,25]. PROG1 appears to function independently from LAZY1. However, it remains unclear that whether PROG1 is also involved in the regulation of polar auxin transport to maintain the prostrate growth in wild rice species. Nevertheless, these findings demonstrated that PROG1 is a major target during rice domestication, which allows the erect growth of the modern rice cultivars [26].
The increased tiller number is usually accompanied by the reduced plant height or vice versa, a phenomenon known as the apical dominance. In crops, plant height is a critical factor in securing potential grain production, which is especially important for lodging-resistance of crops. The Green Revolution in 1960s was marked by the use of semi-dwarf and lodging-resistant varieties of rice and wheat through manipulating biosynthesis and signaling of gibberellin (GA), a phytohormone controlling cell elongation in plants. The biosynthesis and signaling pathways of GA have been well defined. Surprisingly, a kinesin-like protein, encoded by the BRITTLE CULM12/GIBBERELLIN-DEFICIENT DWARF1 (BC12/GDD1) gene was found to bind to the promoter of a key GA synthetic enzyme gene, thereby downregulating its expression [27]. The dwarfism of bc12/gdd1 is also due to altered cellulose microfibril orientation and cell wall composition [28]. Consistent with its dual function as a motor protein and a transcription regulator, BC12/GDD1 is localized in both the nucleus and cytoplasm [27,28]. Plant height in rice is also regulated by the class II formin FH5 that regulates cell expansion and morphogenesis by modulating actin dynamics and the organization of microtubules and microfilaments [29,30].
The mechanical strength of the shoot, a key trait for lodging resistance, is largely attributed to the cell wall, especially the secondary wall. Because cellulose is a predominant component of the cell wall, its synthesis, transport and assembly have attracted considerable attention during the studies related to the mechanical strength. Systematic studies of rice brittle culm (bc) and related mutants, including the above-mentioned bc12, have shed important light on the molecular mechanism of controlling mechanical strength [28,[31][32][33][34][35][36]. The BC genes, along with other genes [37,38], are involved in biosynthesis of cellulose and the cell wall. The BC1 gene encodes a COBRA-like protein that regulates cellulose assembly by interacting with cellulose and affecting microfibril crystallinity [31,36]. BC14 encodes a Golgilocalized nucleotide sugar transporter that plays a critical role in cell wall biosynthesis by transporting UDP-glucose [34].

REGULATION OF LEAF DEVELOPMENT
The leaf architecture, including the size, shape and the angle related to the stem, is a critical factor to determine the photosynthetic efficiency and planting density. Moderate leaf rolling increases photosynthesis capacity. Mutations in SHALLOT-LIKE1/ROLLED LEAF 9 (SLL1/RL9), encoding a MYB-type transcription factor [39,40], cause extremely incurved leaves due to defective programmed cell death (PCD) of abaxial mesophyll cells, resulting in the suppression of specification of abaxial cells [40]. In contrast, the outcurved leaf1 (oul1) mutant shows abaxial leaf rolling. The mutant phenotype is caused by a loss-of-function mutation in a homeodomain leucine zipper transcription gene, which negatively regulates bulliform cell fate and development [41]. Because sll1/rl9 and out1 display a nearly reversed phenotype in specifying of abaxial and adaxial cells, it will be interesting to determine possible genetic interactions between these two loci. In addition, a cellulose 256 National Science Review, 2014, Vol. 1, No. 2 REVIEW synthase-like protein D4 (OsCslD4), a novel protein ABAXIALLY CURLED LEAF1 (ACL1), a putative glycosylphosphatidylinositol-anchored protein SEMIROLLED LEAF1 (SRL1), and a WW domain protein CURLY FLAG LEAF1 (CFL1) have also been found to be involved in the regulation of leaf rolling through different mechanisms [42][43][44][45].
In rice, a major physiological effect of the phytohormone brassinosteroid (BR) is involved in the regulation of leaf angle. LEAF AND TILLER ANGLE INCREASED CONTROLLER (LIC), a CCCH-type zinc finger transcription factor, plays an important role in regulating leaf angle by modulating BR signaling. LIC directly represses the expression of BRASSINAZOLE-RESISTANT 1 (BZR1), a positive regulator of BR signaling, thereby attenuating BR signaling [46,47]. The antagonistic effects of LIC and BZR1 thus illustrate a novel regulatory mechanism of BR signaling to coordinate leaf bending and other developmental events. In contrast to that of LIC, the GRAS-type transcription factor DWARF AND LOW-TILLERING (DLT) is a positive regulator of BR signaling. The dlt mutant displays typical BR-deficient phenotype, characterized as semi-dwarf, reduced tillering and erected leaves (Fig. 1) [48]. DLT is phosphorylated by a GSK3/SHAGGY-like kinase, which is a key regulator of BR signaling [49]. Mutations in TAIHU DWARF1 (TUD1), encoding a U-box E3 ubiquitin ligase, cause a BR-insensitive phenotype. TUD1 is epistatic to D1/RGA1 that encodes the heterotrimeric G protein α subunit [50], and TUD1 protein physically interacts with D1/RGA1, suggesting that they act together to mediate BR signaling [51]. Similarly, a mutation in XIAO, encoding a leucine-rich receptor-like kinase, shows typical BR-deficient phenotype, characterized as dwarfism and reduced leaf angle, and these defects are caused by reduced cell division [52]. The expression of OsGSR1 is induced by GA and repressed by BR. Os-GSR1 acts as an important regulator of the crosstalk of BR and GA signaling [53].
The characterization of INCREASED LEAF ANGLE1 (ILA1), encoding a Raf-Like MAP kinase kinase kinase (MAPKKK), identified a BR-independent mechanism to regulate leaf angle. ILA1 physically interacts with and phosphorylates a small family of novel nuclear proteins [38]. Interestingly, leaf angle is also regulated by a vernalization insensitive 3-like protein encoded by LEAF INCLINATION2 (LC2), whose homologs in Arabidopsis and wheat are involved in regulating flowering through the vernalization pathway [54] (see also below). The NARROW LEAF1 (NAL1) gene was found to regulate leaf blade width and leaf vascular development. NAL1 encodes a novel protein involved in the regulation of polar auxin transport, indicating an important role of auxin in leaf development [55].

ROOT DEVELOPMENT AND NUTRITION
The root system of rice consists of primary roots, lateral roots, adventitious roots (also known as crown roots), and root hairs. The formation of adventitious roots, which are initiated from the stem, is a common feature of grasses. During plant growth, roots directly uptake water and mineral nutrition from the soil and also play a critical role in lodging resistance. Because the rice roots are grown inside the waterflood soil, the study of the root system is technically challenging under natural growth conditions.
Auxin plays a dominant role in the regulation of root development in higher plants, including rice. In recent years, a number of components in auxin signaling were identified in rice and their roles in regulating root development have been functionally characterized [56][57][58][59][60]. A novel protein ROOT AR-CHITECTURE ASSOCIATED1 (OsRAA1) negatively regulates root growth by inhibiting the onset of anaphase [61]. The ORIGIN RECOGNITION COMPLEX (ORC) is a highly conserved component in DNA replication and mutation in OsORC3 causes a temperature-dependent arrest of lateral root development [62].
Considerable efforts have been made to elucidate the molecular mechanism of phosphate (Pi) nutrition. A low Pi level in the soil evokes the Pi-starvation response, characterized as enhanced elongation and root hair growth. OsPHR2, encoding a MYB-type transcription factor and a homolog of Arabidopsis PHR1, regulates a subset of Pi-starvation-inducible genes that are responsible for Pi homeostasis [63]. OsPHR2 directly regulates the expression of the low-affinity Pi transporter gene OsPT2 and this effect is suppressed by a SPX domain-containing protein OsSPX1 [64]. The LEAF TIP NECROSIS1 (LTN1) gene, an ortholog of Arabidopsis PHOS-PHATER ESPONSIVE2 (PHO2), plays a critical role in Pi starvation signaling. The ltn1 mutant shows a typical Pi starvation phenotype, including altered root architecture [65]. The plasma membrane localization of OsPT2 and OsPT8 (a high-affinity Pi transporter) is mediated by PHOSPHATE TRANS-PORTER TRAFFIC FACILITATOR1 (OsPHF1), and the osphf1 mutation renders the retention of these two transporters in the endoplasmic reticulum (ER) [66]. The expression of OsMYB2P-1 is induced by Pi starvation. Overexpression of OsMYB2P-1 causes the altered root architecture and the enhanced tolerance to Pi starvation, associated with the increased expression level of several Pi-responsive or related genes, suggesting that OsMYB2P-1 is an important regulator of Pi starvation signaling [67]. Several rice genes involved in iron homeostasis have also been functionally characterized [68][69][70].

REGULATION OF HEADING AND PANICLE DEVELOPMENT
Flowering time, usually referred to as heading date in crops, is one of the most important agronomic traits that regulate the productivity of rice. Rice is a shortday (SD) plant, namely flowering more rapidly in SD but with delayed flowering under the long day (LD) condition. This photoperiod flowering pathway in plants contains three major components, GIGANTEA (OsGI in rice), CONSTANT (CO; Heading date 1 or Hd1 in rice) and FLOWERING LOCUS T (FT; Heading date 3a or Hd3a in rice).
The OsGI-Hd1-Hd3a route appears to play a limited role in regulating flowering in rice. Instead, a ricespecific pathway dependent on Early heading date 1 (Ehd1) plays a more dominant role in the photoperiod flowering pathway, which is genetically independent on Hd1 (see Fig. 2). Currently available evidence indicates that Ehd1 is a key convergence point regulated by multiple input signals or signaling pathways. Several regulators in the Ehd1 pathway have been characterized. Transition from the vegetative phase to reproductive phase is a perquisite for flowering or heading. Mutations in the RICE INDETERMI-NATE 1 (RID1) gene shows a never-flowering phenotype. RID1, encoding a Cys-2/His-2-type zinc finger transcription factor that have no close homologs in Arabidopsis, positively regulates the expression of the genes in flowering regulation, especially in the Ehd1-Hd3a pathway, indicating that RID1 acts upstream of the photoperiod pathway to regulate the vegetative to reproductive phase transition and flowering [71]. Similar results were also obtained by two independent studies [72,73].
The quantitative trait locus (QTL) Ghd7 (grain number, plant height and heading date) has major effects on several important traits as implied from its name. Enhanced expression of Ghd7, encoding a CCT domain protein, causes delayed heading and increased plant height and panicle size. Ghd7 represses the expression of Hd3a in LD, possibly through Ehd1, in the photoperiod flowering pathway. Association analysis of allelic variants suggests that Ghd7 plays an important role for both productivity and adaptability of rice globally [74]. This adaptability has also been found in the DTH2 (Days To Heading on Chromosome 2) and Ghd7.1 genes. DTH2 encodes a CO-like protein that induces the expression of Hd3a and RICE FLOWERING LOCUS T 1 (RFT1) independent of Ehd1 and Hd1. Remarkably, two nucleotide polymorphisms in DTH2 were found to be correlated with the northward expansion of rice cultivation under natural LD conditions in Northern Asia, suggesting that DTH2 represents an important target of human selection during domestication [75]. Ghd7.1, allelic to Hd2 and encoding a pseudo-response regulator (Os-PRR37), appears to be strongly selected during domestication. Multiple Ghd7.1 alleles were identified in different cultivars, of which a strong allele and a non-functional allele are widely distributed in various cultivars. The strong allele, which causes delayed heading under LD, is usually presented in singlecropping varieties, with a greater yield potential and grown in central and Southern China and Southeast Asia. On the other hand, the non-functional allele is mainly found in the early-season rice cultivars. Together with analysis of other alleles, it was concluded that different Ghd7.1 alleles have distinct eco-geographical distribution patterns [76]. Consistent with these observations, Ghd7.1 represses the expression of Ehd1 and Hd3a under LD [76].
The ehd4 mutant shows a significantly late flowering phenotype under both LD and SD conditions. Ehd4 was characterized as a novel CCCHtype zinc finger protein that positively regulates the expression of Ehd1, Hd3a and RFT1 in an Ehd1dependent manner. Interestingly, Ehd4 is highly conserved in genus Oryza, but is absent in any other species, suggesting that Ehd4 is Oryza-specific flowering regulator [77]. Conversely, DTH8, also known as Ghd8/LHD1/Hd5, encoding a putative HAP3 subunit of the CCAAT-box-binding transcription factor, negatively regulates flowering by repressing 258 National Science Review, 2014, Vol. 1, No. 2 REVIEW the expression of Ehd1 and Hd3a (Fig. 2). Both the DTH8 and Ghd8 alleles show similar phenotypes as Ghd7 and Ghd7.1 in heading, plant height and yield-related traits [78][79][80]. Interestingly, the expression level of MOC1 in the panicle is significantly higher in a near-isogenic line (NIL) carrying a functional DTH8/Ghd8/LHD1/Hd5 gene than that with a nonfunctional gene, suggesting that DTH8/Ghd8/LHD1/Hd5 may act upstream of MOC1 to regulate panicle development [79].
In addition to the photoperiod pathway, the phytohormone GA also plays a critical role in the regulation of flowering. The earlier flowering1 (el1) mutant displays an enhanced GA response. EL1 encodes a casein kinase I that specifically phosphorylates the DELLA protein SLR1, a repressor of GA signaling, thus negatively regulating the activity and stability of SLR1 [81]. A mutation in the rice ethylene receptor ETR2 gene causes early flowering, whereas overexpression of ETR2 delays floral transition. The etr2 mutant phenotype is associated with a decreased expression level of several genes in the photoperiod pathway, including OsGI and TERMINAL FLOWER1/CENTRORADIALIS homologs [82]. Although rice is a non-vernalized species, mutations in LC2, encoding a vernalization insensitive 3-like protein, causes a late flowering phenotype, suggesting that heading is also somehow regulated by a vernalization pathway [54]. The late flowering phenotype and enlarged leaf angles of the lc2 mutant may be partly attributed to a pleiotropic effect.
The specification of floral organ identity is one of best models in developmental biology and has drawn extensive attentions in decades. Studies in Arabidopsis and snapdragon have established the ABCDE model-based framework of flower development, which appears to be conserved in higher plants. A great deal of efforts have been made to functionally characterize many rice genes involved in specifying rice floral organ identities [83][84][85]. However, grasses, including rice, have unique floral patterns, which are regulated by rice-specific mechanisms. Rice flowers have unique outer floral organs, such as palea and glume. Mutations in the EXTRA GLUME1 (EG1) gene, encoding a putative lipase, cause misspecified fates of the glume and the floral meristem [86]. The chimeric floral organs1 (cfo1) mutant shows multiple defects in flower development, including impaired marginal regions of the palea and chimeric floral organs. CFO1 encodes a monocotyledon-specific MADS transcription factor that negatively regulates the expression of DROOP-ING LEAF, a floral organ identity gene [87].
Pseudovivipary is an important reproductive strategy in some grasses, in which asexual propag-ules are formed in place of sexual reproductive structures. The underpinning molecular mechanism of pseudovivipary remains completely unknown. The analysis of a pseudovivipary mutant phoenix (pho) revealed that pho carries mutations in two independent loci, DEGENERATIVE PALEA/OsMADS15 and ABNORMAL FLORAL ORGANS/OsMADS1, indicating that these two transcription factors are required to ensure sexual reproduction. This study identifies a novel mechanism regulating reproductive habit in rice [88].
Grain yield in rice is determined by three main components, panicle number per plant, grain number per panicle and mean grain weight [89]. The moc1 mutations not only cause impaired tillering, but also result in the reduced panicle branching number and spikelets, indicating that MOC1 plays an important role in the regulation of branching in both vegetative and reproductive developmental stages [3,11]. As mentioned above, this notion is supported by the observation that DTH8/Ghd8 promotes the MOC1 expression to regulate panicle branching [79]. The selection of optimal panicle architecture, defined as dense and erect panicles, is demanding for improving rice productivity. DENSE AND ERECT PANICLE1 (DEP1), a major QTL responsible for panicle architecture, was mapped and cloned, which encodes a PEBP (phosphatidylethanolamine-binding protein) like domain protein. The dep1 gain-of-function mutant allele carries a premature stop codon mutation, resulting in the formation of a truncated protein. The dep1 allele, widely presented in many high yield rice varieties, enhances the meristematic activity, characterized by the formation of a larger shoot apex meristem (SAM) and a greater number of panicle meristems compared with the wild type DEP1 allele. As a result, the dep1 allele causes a reduced length of the inflorescence internode and an increased number of grains per panicle, leading to a substantially increased grain yield. Interestingly, several C-terminal truncated deletions were found in barley and bread wheat and the truncated wheat dep1 allele functions similarly as its rice homolog, suggesting that a conserved mechanism operates in cereals [90]. DEP1 is allelic to qPE9-1, a QTL locus identified in an independent study, and these two alleles shows nearly identical phenotypes [91]. The expression of DEP1 is positively regulated by IPA1 that directly binds to the DEP1 promoter, consistent with the role of IPA1 in regulating panicle development [21]. The dense and erect panicle phenotype is also regulated by ERECT PANI-CLE2 (EP2; also known as DEP2), an ER-localized protein with unknown function [92,93]. However, it remains unknown whether EP2/DEP2 and REVIEW Zuo and Li 259 DEP1 act in a same pathway to regulate panicle development.
In a previous study, Gn1a, encoding a cytokinin oxidase (OsCKX2) catalyzing irreversibly degradation of the phytohormone cytokinin, was identified as a major QTL controlling panicle branching and grain number by modulating the level of cytokinin that promotes meristematic activity [94]. In the NIL carrying the dep1 allele (NIL-dep1), the expression of Gn1a/OsCKX2 was substantially reduced than that in NIL-DEP1 plants [90], implying that DEP1and Gn1a/OsCKX2-regulated panicle development may partly share a similar mechanism. Recently, DROUGHT AND SALT TOLERANCE (DST) was identified as a direct transcription regulator of Gn1a/OsCKX2. DST encodes a zinc finger type transcription factor that directly regulates the expression of Gn1a/OsCKX2 in the reproductive meristem. A semi-dominant negative mutation in DST, designated as regulator of Gn1a 1 (DST reg1 ), impairs the DST-directed expression of Gn1a/OsCKX2 and thus increases the cytokinin level in the reproductive meristems, resulting in the enhanced panicle branching and the increased grain number. Similarly, the expression of a DST reg1 transgene in wheat causes an increased ear size and spikelet number [95]. Interestingly, the loss-of-function mutations in DST also cause the increased panicle number with a reduction of the panicle length, significantly increased leaf width, and resistance to abiotic stresses [96] (see also below), suggesting that DST regulates a wide range of physiological processes in rice growth and development. The LAX2 gene regulates branching of most aerial parts. In lax2, the formation of the axillary meristems is absent in most of the lateral branching of the panicle, leading to no secondary branch formation in the panicle. LAX2 genetically interacts with MOC1 and LAX1 to regulate axillary meristem formation [11]. Although LAX1, LAX2 and DEP1 are all involved in the secondary branch formation, no significant difference of the LAX1 expression was found between the NIL lines carrying DEP1 and dep1 alleles, suggestive of the involvement of different mechanisms [90].
Panicle size is a critical factor for grain yield [89]. Enhanced expression of either Ghd7 or IPA1 increases the panicle size, indicating that these two genes are positive regulator of panicle growth [19,20,74]. Mutations in the SHORT PANICLE1 (SP1) gene cause the reduction of the panicle size. SP1 encodes a putative membrane-localized protein belonging to the PTR family transporters that are involved in the transport a wide range of nitrogencontaining substrates. SP1 positively regulates panicle growth, possibly acting in transport of nitrogen or other unidentified compounds [97]. Arginine is an important amino acid for the transport and storage of nitrogen in plants. A mutation in the OsARG gene, encoding an arginine hydrolysis enzyme arginase, causes the reduction in panicle size and grain size. Overexpression of OsARG increased grain number under nitrogen-limited conditions, indicating that the homeostasis of arginine plays an important role in panicle development [98]. The molecular mechanism regulating the panicle size remains largely unknown.

REGULATION OF REPRODUCTIVE DEVELOPMENT AND FERTILITY
Meiosis is a specialized form of cell division during sexual reproduction of eukaryotes, which allows the formation of haploid gametophytes from the diploid parental cells. Meiosis is under the tight genetic control and is mechanistically conserved in eukaryotes. Although the major regulatory components of meiosis have been well characterized in yeast and animal cells, their homologous proteins in higher plants are poorly understood. In recent years, several rice meiosis-specific proteins have been functionally characterized. In meiosis, homologous recombination is initiated by programmed DNA double-strand break (DSB) and the assembly of the synaptonemal complex (SC). CENTRAL REGION COM-PONENT1 (CRC1), an important component of SC, is essential for meiotic DSB formation and for the recruitment of HOMOLOGOUS PAIRING ABERRATION IN RICE MEIOSIS2 (PAIR2) onto meiotic chromosomes [99]. During later stages, whereas PAIR3, encoding a novel protein, is required for homologous chromosome pairing and synapsis [100,101], MER3 is required for meiotic crossover formation in a PAR2-dependent manner [102]. Both ZEP1 and ZIP4 function as the central elements of SC to regulate crossovers [103,104], and SC, in turn, is stabilized by OsSGO1 that is also involved in the protection of centromeric cohesion [105]. Moreover, HEI10, a novel RING domain protein that is colocalized with ZEP1 and partly colocalized with MER3, is likely involved in the formation of interference-sensitive crossovers [106]. Notably, the centromeric recruitment of OsSGO1 is dependent on BRK1, a Bub1-related kinase that regulates the tension across homologous kinetochores at metaphase I [107]. During meiosis, OsCOM1 appears to play a more general role, as the mutations in OsCOM1 cause severe defects in SC assembly, homologous pairing and recombination [108]. Despite extensive genetic studies, very little is known about the biochemical functions of these proteins in meiosis.

REVIEW
In recent years, a large body of evidences has been obtained on the regulatory mechanism of male gametophyte development in rice by characterizing various male sterile mutants, whereas relatively fewer efforts have been made in investigating female gametophyte development. Normal development of pollen requires proper sugar partitioning, a process of transport of carbohydrates from photosynthetic tissues (sources) to sink tissues for nutrition and energy. In these processes, a key regulator is a MYB transcription factor CAR-BON STARVED ANTHER (CSA) [109]. As a photoperiod-controlled male sterile mutant, csa is promising for the establishment of a stable twoline hybrid system [110]. A panicle-dominantly expressed gene, encoding a kinesin-1-like motor protein POLLEN SEMI-STERILITY1 (PSS1), is presumably involved in microtubule organization, and plays an essential role in male meiotic chromosomal dynamics, male gametogenesis and anther dehiscence [111]. Lipidic compounds are important components of the anther cuticle and pollen exine. Several studies reveal the importance of lipid metabolism and cutin biosynthesis in the formation of the anther cuticle and pollen exine [112][113][114]. In addition, a fine-tuned balance of reactive oxygen species and the cellular redox status also play important roles in normal development of anthers and pollen [115,116]. During pollen maturation, three MIKC * -type MADS transcription factors play important and partially redundant roles by forming heterodimers to regulate downstream target genes [117].
In higher plants, the entire development process of male gametophytes is completed in the anther, in which tapetal cells form the innermost layer of the anther. Upon the completion of meiosis, the haploid male gametophyte undergoes two cycles of mitosis and is eventually developed into mature pollen, which is then released from the anther. During this process, tapetal cells are degenerated through PCD. Premature or delayed degradation of tapetal cells causes defective pollen development or male sterility. Two helix-loop-helix transcription factors TAPETUM DEGENERATION RETAR-DATION (TDR) and ETERNAL TAPETUM 1 (EAT1) were found as key regulators that control a signaling cascade promoting tapetal cell death. TDR and EAT1 directly regulate a cysteine protease gene (OsCP1) and two aspartic protease genes (OsAP25 and OsAP37; homologs of OsAP25 and OsAP37 have been known to induce PCD in both yeast and plants), respectively. Whereas TDR genetically acts upstream of EAT1, both proteins are present in a complex which may regulate common targets [118,119]. The involvement of apoptotic type cell death in tapetal cell degradation is further evident by characterizing APOPTOSIS INHIBITOR5 (API5), a homolog of antiapoptosis protein Api5 in animals. A mutation in API5 results in delayed degeneration of tapetal cells. API5 physically interacts with AIP1 and AIP2, which form a dimer and may directly regulate of OsCP1, a target of TDR [120]. Although it appears that TDR, EAT1 and API5 may regulate a same subset of target genes, it however remains unknown whether API5 acts independently from TDR and EAT1 or in a linear pathway with the latter two transcription factor genes.
It has long been suspected that the formation of the haploid male gametophytes requires active information exchanges between male gametophytes and the surrounding somatic tissues. A recent study identified a secretory fasciclin glycoprotein MICROSPORE AND TAPETUM REGULA-TOR1 (MTR1) that plays a critical role in mediating the interaction between sporophytic and reproductive cells. Mutations in MTR1 impair development in both male gametophytes and the surrounding tapetum. MTR1 protein is secreted from male reproductive cells, thereby regulating the development of reproductive cells and their adjacent somatic cells [121]. This study reveals a novel regulatory mechanism of plant male reproductive development.
The cytoplasmic male sterility (CMS) phenotype is widely found in higher plants, which is caused by the incompatibilities between the organellar and nuclear genomes, thus preventing self-pollination but favoring cross-pollination. Currently, the generation of hybrid crops is largely based on various CMS lines. In China, the Wild Abortive CMS (CMS-WA) is the most widely used line in three-line hybrid rice production since 1970s. Yao-Guang Liu and coworkers identified a new mitochondrial gene WA352 that was recently originated in wild rice. The WA352 protein interacts with COX11, a mitochondria-localized protein encoded by the nuclear genome, to inhibit the COX11 activity in scavenging peroxide, thereby triggering premature tapetal cell death and defective pollen development. Moreover, the sterility in the CMS-WA line can be restored by two nuclear restorer-of-fertility (Rf ) Rf3 and Rf4, which likely suppress the WA352 function posttranslationally and posttranscriptionally, respectively (Fig. 3). Discovery made in this study uncovered the molecular mechanism of CMS-WA, by which a detrimental interaction between a newly evolved mitochondrial gene and an essential nuclear gene causes CMS [122].
The CMS phenotype can often be rescued by nuclear Rf genes, which are essential for the practice of three-and two-line hybrid rice production. The Hong-Lian CMS is an important line in three-line hybrid rice production, and CMS in this line is partly attributed to the altered mitochondrial protein assembly, the impaired jasmonic acid biosynthesis and the mitochondrial gene atp6-orfH79 [123,124]. In particular, ORFH79 protein interacts with a subunit of the mitochondrial electron transport chain complex III to inhibit the enzyme activity of complex III [124]. On the other hand, RF5, a pentatricopeptide repeat protein encoded by the Rf5 gene for Hong-Lian CMS, forms a complex with a Gly-rich protein GRP162, which binds to and subsequently cleave the atp6-orfH79 transcript, thereby restoring fertility (Fig. 3) [125].
Sterility is also known as a common consequence in hybrids, and this reproductive barrier or incompatibility restricts gene flow among divergent populations. The fertilities of the hybrids between indica and japonica are usually severely reduced, mainly due to postzygotic reproductive isolation. However, some particular rice varieties are capable of producing highly fertile hybrids when crossed either to indica or japonica, a phenomenon known as wide-compatibility. Several genes for indica-japonica hybrid sterility have been cloned and analyzed. The S5 locus was proposed as a major regulator for indica-japonica hybrid fertility, and consisted of three major alleles, including an indica allele (S5-i), a japonica allele (S5-j), and a neutral allele (S5-n), which is referred to as the wide compatibility gene. When crossing plants carrying S5-n with plants carrying either S5-i or S5-j, the hybrid will be fertile, whereas the combination of S5-i and S5-j causes abortive embryo-sacs. The S5 locus contains 5 ORFs (open reading frame; ORF1 through ORF5), of which ORF5 encodes a putative aspartic protease mainly localized in the cell wall. The ORF5 of S5-i and S5-j alleles differ by two nucleotides and the predicted proteins carry variations in two amino acid residues. The ORF5 of S5-n allele contains a deletion, leading to the formation of a truncated protein lacking the N-terminus that mislocalized in the cytoplasm [126]. This triallelic system appears to occur independently in indica and japonica, and the allele-specific signatures are tightly associated with the hybrid sterility [127]. Further studies revealed that ORF3 and ORF4 encode a HSP70-like protein and a novel transmembrane protein, respectively. Similar to that ORF5, both ORF3 and ORF4 also carry mutations rendering the encoded protein nonfunctional in various varieties (referred to as ORF+ and ORF-for functional and nonfunctional alleles, respectively, for each of these three genes). On the basis of the genetic and molecular studies, a killer-protector model was proposed to REVIEW explain the S5-regulated hybrid sterility [128]. During female sporogenesis, ORF5+ (killer) functions together with ORF4+ (partner) to induce the ER stress, which is antagonized by ORF3+ (protector) and thus to generate normal gametes. In the ORF3-background, the ER stress evokes premature cell death, eventually resulting in embryo-sac abortion.
The Sa locus was found as a major regulator of indica-japonica hybrid male sterility. The Sa locus consists of two adjacent genes, SaM and SaF. Whereas most indica cultivars contain a haplotype SaM(+)SaF(+), all japonica cultivars have a SaM(−)SaF(−) genotype. The SaM(−) encodes a truncated protein lacking an inhibitory domain that prevents the interaction between the full-length SaM(+) and SaF(+). In indica-japonica hybrids (heterozygous at the SaM SaF locus), pollen carrying a SaM(−) allele are aborted due to the interaction of its encoded protein with SaF(+), leading to male semi-sterility, and this male sterility also requires the SaM(+) allele. Hybrids lacking any of the three alleles will be male fertile. Therefore, the interactions among the three components derived from two genes regulate hybrid male sterility [129].
Photoperiod-sensitive genic male sterility (PGMS) is referred to as the fertility regulated by day length, which is extremely useful for the practice of two-line hybrid rice production. A spontaneous PGMS mutant 58S in a japonica variety Nongken 58 (58N) was found 40 years ago and has been successfully used in hybrid rice breeding since then. However, despite extensive studies, very litter is known about the molecular mechanism of the PGMS phenotype of 58S. Recent studies reveal that the PGMS phenotype of 58S is caused by the reduced expression level of a long noncoding RNA, referred to as long-day-specific male-fertility-associated RNA (LDMAR). A SNP (Single Nucleotide Polymorphism) between 58N and 58S in LDMAR results in increased methylation of the putative promoter of LDMAR in 58S, which reduces the expression of LDMAR and subsequently induces PCD in the anther under long-day conditions [130][131][132]. Revealing of this novel regulatory mechanism on PGMS has important implications in hybrid crop breeding.
Heterosis presented in hybrids is one of the most important scientific questions remained to be addressed in centuries, yet the underpinning genetic basis and molecular mechanisms are largely unknown. Using a unique "immortalized F 2 " population derived from an elite rice hybrid, the genetic composition of several key yield traits was investigated. The relative contributions of the genetic com-ponents were found to vary significantly with specific traits, suggesting that the cumulative effects of these components may play a key role to determine the genetic basis of heterosis in the hybrid [133].
The superhybrid rice Liang-You-Pei-Jiu (LYP9) is one of the most popular varieties grown in China and other Asia-Pacific areas. LYP9 was developed by crossing the paternal 93-11 variety (indica) and the maternal PA64s variety (mixed genetic background of indica and javanica). In a transcriptomic analysis of LYP9 and its parents, it was found that genes related to energy metabolism and transport are substantially enriched in differentially expressed genes between the hybrid and its parents, suggesting that these genes may have important contributions to heterosis in the hybrid [134]. By resequencing of 132 recombinant inbred lines (RILs) of LYP9 and its parental lines, a high-resolution linkage map was constructed [135]. Based on this high quality map, the genome sequences of the parental lines were significantly improved and 43 yield-associated QTLs were detected. In particular, DTH8 and LAX1 were identified as candidate genes for two QTLs, qSN8 and qSPB1, respectively. A genetic complementation test demonstrated that DTH8 indeed represents qSN8 [135]. This study provides a promising strategy to dissect QTLs associated with complex agronomic traits.

REGULATION OF SEED DEVELOPMENT, SEED SHATTERING AND SEED COAT FORMATION
Grain size and shape are important determinants of grain yield and quality in cereal crops, which are usually controlled by QTLs. In rice, hundreds of QTLs for grain size (GS), grain length (GL), and grain width (GW) have been found, and a few of them have been cloned and molecularly characterized in some details, which will have great impacts on breeding [89,136].
GS3, a major QTL for both grain weight and grain length, encodes a novel protein containing a plant-specific organ size regulation (OSR) domain, a transmembrane domain, a tumor necrosis factor receptor/nerve growth factor receptor (TNFR/NGFR) family cysteine-rich domain, and a von Willebrand factor type C (VWFC) domain. A nonsense mutation was identified in all the examined large-grain varieties, suggesting that GS3 is a negative regulator of grain size and has been subjected to strong selection in breeding [137,138]. A detailed analysis of the functional domains of GS3 protein identified the OSR domain as the negative regulatory motif, whose activity, in turn, is repressed by the TNFR/NGFR and VWFC REVIEW Zuo and Li 263 domains. Consistent with the loss of the OSR activity resulting in long grains, loss-of-function mutations in the TNFR/NGFR and VWFC domains produce short grains [139]. Interestingly, GS3 shares some homology with DEP1 at the N-terminal region and the VWFC domain, implying a possible function similarity. Indeed, DEP1 also negatively regulates grain size, a role similar to that of GS3 [90,91]. GS5 was identified as a putative serine carboxypeptidase that acts a positive regulator of grain size. Several lines of evidences, including transgenic studies and association analysis, demonstrate that the GS5promoted larger grain size is caused by polymorphisms in the GS5 promoter, which may result in different expression levels of GS5, thus regulating grain size [140]. Three QTL loci mainly controlling seed width have been cloned by Chinese scientists. GW2, encoding a RING-type E3 ubiquitin ligase, negatively regulates cell division via unidentified substrates. The loss-of-function mutations in GW2 result in the increased cell number and consequently wider seeds. Meanwhile, gw2 also enhances the grain milk filling rate, resulting in the increased grain weight. Together, these traits lead to a higher grain yield [141]. GW5 encodes a novel nuclear protein localized in the nucleus. Molecular and association analyses revealed that a large deletion (∼1.2 Kb) in the GW5 gene was tightly associated with the grain-width phenotype, suggesting that GW5 negatively regulates grain width and the gw5 mutation is an important target during domestication. Interestingly, polyubiquitin was identified as an interacting protein of GW5, implying that GW5 may act in the proteasomal degradation pathway to regulate grain size [142]. In contrast to that of GW2 and GW5, GW8, encoding a putative transcription factor OsSPL16, acts as a positive regulator for grain size and grain milk filling rate by promoting cell division. Similar to that of GS5, the promoter region of GW8 has been likely selected during breeding [143].
Thus far, the only identified major QTL of grain length is qGL3/qGL3.1, which contributes significantly to grain thickness and grain width [144][145][146]. qGL3/qGL3.1 encodes a putative protein phosphatase with Kelch-like repeat domain (OsPPKL1) that directly dephosphorylates its substrate, Cyclin-T1;3 to regulate cell division. Genetic and transgenic studies revealed that qGL3/qGL3.1 acts as a negative regulator of grain length. Analysis of 94 rice germplasms indicated that the qgl3 is a rare allele that has not been selected in breeding, a case similar to that of ipa1 and gw2 alleles [19,141]. Thus, the rareness of the qgl3/qgl3.1 allele offers great potential in rice breeding [144].
Grain filling is a critical trait, which directly determines grain weight. During this process, carbohydrates synthesized in the photosynthetic sources are transported into the sink (grains), which is correlated with the switch of the metabolic flow in filling grains from central carbon metabolism to alcoholic fermentation [147]. A plastidial UDPglucose epimerase (UGE) involved in galactolipid biosynthesis is required for chloroplast biogenesis and photosynthetic activity. Mutations in the related gene PHOTOASSIMILATE DEFECTIVE1 (PHD1) cause altered metabolic profiling of galactolipids and decreased photosynthetic activity, whereas overexpression of PHD1 increases photosynthetic efficiency and grain production [148]. The GRAIN INCOMPLETE FILLING 1 (GIF1) encodes a cell-wall invertase involved carbon partitioning during grain filling, which appears to be selected during domestication [149]. Interestingly, several key events in the developing endosperm, including cell division, cell growth and starch accumulation, are diurnally regulated. In particular, starch synthesis proteins are downregulated by light but upregulated in the dark, a phase opposite to the diurnal cycle of photosynthesis, illustrating highly dynamic and coordinated cellular and metabolic processes in endosperm development [150]. After fertilization, maternal tissues are degenerated through PCD in developing seeds. The knockdown of the expression of OsMADS29 results in abnormal seed development caused by defective PCD in maternal tissues [151].
Seed awns, seed shattering and seed coat (or seed hull in rice) color are important agricultural traits in crop domestication. In wild rice, the formation of long awns is critical for seed dispersal, but is also an energy-consuming process, leading to the reduced grain productivity. A recent study identified a major QTL locus Awn-1 (An-1) as a key regulator for long awn formation in wild rice O. rufipogon. An-1, encoding a basic helix-loop-helix transcription factor, positively regulates awn elongation and negatively regulates the grain number, and is a major target for artificial selection in cultivated rice [152].
During rice domestication, the elimination of seed shattering in wild rice is also a key step, which allows easy harvest and a reduced loss in production for cultivated rice. This trait is controlled by SHAT-TERING1 (SHA1; also known as SH4 for GRAIN SHATTERING QTL ON CHROMOSOME 4), encoding a plant-specific trihelix transcription factor. A single amino acid substitution caused by a single nucleotide change in SHA1 in all examined 200 cultivated rice in both indica and japonica cultivars abolishes the seed shattering trait [153,154]. SHA1 positively regulates the expression of a downstream REVIEW target, SHATTERING ABORTION1 (SHAT1) that encodes an AP2-type transcription factor and is required for the specification of the abscission zone [155].
During grain ripening, the seed hull of wild rice (e.g., O. rufipogon) is black, whereas that of cultivated rice is straw-white. In O. rufipogon, this trait is controlled by the Black hull4 (Bh4) gene that encodes an amino acid transporter. Similar to that of SHA1, the Bh4 gene in all examined cultivars carries various loss-of-function mutations that cause the formation of straw-white seed hull, suggesting strong selection during domestication. The strawwhite seed hull trait was proposed to be selected as a visual phenotype of nonshattered grains during rice domestication [156]. In maturing and mature seeds, pre-harvest sprouting (PHS) or vivipary frequently occurs, which affects both the yield and the quality of the grain in cereals. The characterization of 4 rice phs mutants revealed that their wild type alleles encode enzymes for abscisic acid (ABA) biosynthesis, and these mutations cause a reduced level of ABA, leading to the PHS phenotype [157].

CONTROL OF GRAIN QUALITY
Tremendous efforts have been made to increase grain yield of major crops in breeding, but relatively less efforts given to the improvement of grain quality. Grain quality has now become the primary target of rice consumers and breeders. In rice, eating and cooking quality is mainly determined by three physicochemical factors, namely amylose content (AC), gel consistency (GC) and gelatinization temperature (GT). The endosperm AC is well correlated with the pre-mRNA splicing patterns of Waxy (Wx), which encodes a starch synthase, in various cultivars [158], whereas GT is mainly affected by the ALK gene, encoding soluble starch synthase II (SSII-3) [159]. The dull endosperm of rice grains is a classical morphological and agronomical trait that has long been exploited for breeding and genetic studies. Dull endosperm1 (Du1), encoding a member of pre-mRNA processing (Prp1) family, is mainly expressed in panicles, specifically affects the splicing efficiency of japonica Wx transcripts, and regulates starch biosynthesis by mediating the expression of starch biosynthesis genes [160].
In an association analysis, the correlation of 18 starch synthesis-related genes with the eating and cooking quality was examined [161]. Waxy shows the highest correlation with AC and GC, and ALK is the major contributor of GT. Moreover, ALK also show significant correlations to GC and AC, and Waxy to GT, respectively. A number of minor effect genes have also been identified for each prop-erty. Notably, the actions of these genes are correlated each other, thereby forming a fine tuned network, which controls the eating and cooking quality by coordinating these three properties. Major findings made in the association analysis were verified by transgenic studies [161].
An AP2 transcription factor RICE STARCH REGULATOR1 (RSR1), identified by coexpression analysis, negatively regulates the expression of starch synthesis genes in seeds. The rsr1 mutant seeds show an increase in AC, a reduction in GT, and producing larger seeds [162]. In a large scale of genetic screen of seed quality mutants, over 550 mutants were identified, which affected the contents of starch, amylose, protein and lipids [163].
Fragrance is an important parameter of rice grain quality. Genetic analysis shows that a single recessive mutation (fgr) is associated with rice fragrance and its dominant Fgr allele is associated with the lack of fragrance. The FGR gene was identified to encode betaine aldehyde dehydrogenase (BADH2) catalyzing the oxidization of 4-aminobutyraldehyde (ABald) that is presumed as the precursor of 2-acetyl-1pyrroline (2AP), a potent flavor component in rice fragrance. The fgr/badh2 mutations cause the accumulation of AB-ald that enhances 2AP biosynthesis, thus producing fragrant grains [164].

RESISTANCE TO DISEASE AND INSECT PESTS
Rice bacterial blight and fungal blast and sheath blight are major rice diseases in China, often causing heave and even devastating loss of rice yield. Battles against rice diseases have long been a strenuous task for breeding, yet the molecular mechanism of the rice-pathogen interaction is poorly understood. Bacterial blight caused by Xanthomonas oryzae pv. oryzae (Xoo) is one of the most dangerous rice diseases. In the rice-Xoo interaction, OsMPK6 functions as an activator for local resistance to Xoo, but acts as a repressor for systemic acquired resistance in systemic tissues, involving in salicylic acid (SA) and jasmonic acid (JA) signaling in both cases [165]. Mutations in C3H12, encoding a CCCHtype zinc finger protein and colocalized with a minor disease resistance QTL to Xoo, causes the increased susceptibility to Xoo, whereas overexpression of C3H12 enhanced resistance to Xoo, accompanied by altered JA signaling [166]. Expression of the resistance (R) gene XA13 is induced by Xoo infection, which is required for bacterial growth. Mutations in the XA13 promoter (the xa13 allele) reduce its expression induced by Xoo, thus conferring disease resistance. Unexpectedly, XA13, encoding a novel transmembrane protein, is also REVIEW Zuo and Li 265 required for pollen development, thus linking these two separate biological processes together [167]. The XA13 protein physically interacts with COPT1 and COPT5, two copper transporters, to mediate the removal of copper from xylem vessels where Xoo propagates and spreads to cause disease. Because copper suppresses Xoo growth, the redistribution of copper mediated by the XA13-COPT complex, which is activated by Xoo infection, facilitates Xoo growth. These studies illustrate a novel strategy employed by Xoo to overcome the defense system of rice [168]. When challenged by unfavorable conditions such as pathogens and abiotic stresses, plants usually respond by slowing down or even stopping growth. As such a mechanism, the expression of GH3-8, belonging to a small family genes encoding indole-3-acetic acid (IAA)-amino acid synthetases, is induced by pathogens. Free IAA is a major biologically active form of auxin that plays an essential role in promoting plant growth and development, whereas the formation of the conjugated IAA-amino acids results in the reduced level of free IAA. Overexpression of GH3-8 or GH3-2 causes enhanced resistance to Xoo and other pathogens, independent of JA and SA signaling, accompanied by retarded plant growth and development [169,170]. Moreover, overexpression of a CNB-LRR-type R gene NLS1 in two semi-dominant allelic mutants nls1-1D and nls1-2D causes resistance to Xoo independent of SA signaling [171]. Two adjacent NBS-LRR-type R genes, PIK-1 and PIK-2, are required for the resistance to rice blast, which appear to emerge after rice domestication [172]. Plant innate immunity is dependent on the detection of the microbe-associated molecular patterns of pathogens by pattern recognition receptors (PRR) localized at the plant cell surface. Two lysin motifcontaining proteins, LYP4 and LYP6, were characterized as functional PRRs to specifically recognize and bind both bacterial peptidoglycan (PGN) and fungal chitin. Knockdown of the expression of either LYP specifically causes the reduced defense response induced by PGN or chitin, and compromises the resistance against both bacterial pathogen Xoo and fungal pathogen Magnaporthe oryzae [173]. A putative E3 ubiquitin ligase OsBBI1 is involved in the broad-spectrum resistance against Magnaporthe oryzae by modulating the reactive oxygen species (ROS) level and the cell wall structure [174].
The brown planthopper (BPH) is the most destructive pest on rice by ingesting phloem sap of the plant. Although multiple BPH-resistant loci have been identified in various rice varieties and some of these loci have been used in breeding, none of the related genes have been molecularly characterized till recently. BPH14, which confers brown planthopper-resistance at seedling and mature stages, was recently identified through positional cloning. BPH14 encodes a CC-NB-LRR protein containing a unique LRR domain and is predominantly expressed in vascular bundles where planthopper ingests on the plant. The BPH14-mediated defense against brown planthopper is dependent on SA signaling and is involved in callose deposition in phloem cells and trypsin inhibitor production [175,176].

TOLERANCE TO ABIOTIC STRESSES
The most influential abiotic factors affecting rice yield are salinity, cold and heat stresses, and many genes modulating stress tolerance have been identified and characterized. Several major QTLs for Na + and K + uptake in shoots and roots have been identified [177], of which SHOOT K + CONCENTRA-TION1 (SKC1) was molecularly characterized to encode a Na + -selective transporter involved in the regulation of K + /Na + homeostasis. Between saltresistance and salt-susceptible varieties, four aminoacid substitutions were found to be located in the loop regions of the putative transmembrane domains of SKC1, implying their functional importance [178]. Stomata on the epidermis of leaves and other organs control gas exchange, such as uptake of carbon dioxide and oxygen and evaporation of water, thereby playing crucial roles in photosynthesis, respiration and abiotic stress tolerance. The zinc finger transcription factor DST negatively regulates stomatal closure through directly targeting several genes involved in hydrogen peroxide homeostasis. The dst mutation causes the increased stomatal closure and reduces stomatal density, thereby conferring the tolerance to drought and salt [96]. DST has also been found to directly modulate the level of cytokinin [95], a phytohormone also implied to play an important role in regulating the stress response. It will be interesting to explore whether the DSTregulated stress response is involved in cytokinin signaling.
The expression of OsSKIPa, encoding a conserved protein with a SNW/SKIP domain, is induced by various abiotic stresses and several stress-related phytohormones. Overexpression of OsSKIPa results in tolerance to various stresses, accompanied with the elevated expression of stressrelated genes [179]. Similarly, the expression of a MAPKKK gene DROUGHT-HYPERSENSITIVE MUTANT1 (DSM1) and a receptor-like kinase gene OsSIK1 is induced by various stresses, and overexpression of these genes results in the improved tolerance to drought and salt, associated with the increased activities of ROS scavenging enzymes, whereas the knockout or reduced expression REVIEW of these genes causes a stress-sensitive phenotype [180,181].
The phytohormone ABA is a master regulator of the stress response and ABA-mediated stress signaling pathways have been extensively studied. The DSM2 gene encodes a β-carotene hydroxylase that catalyzes the formation of zeaxanthin, a carotenoid precursor of ABA. Mutations in DSM2 cause a significantly reduced level of zeaxanthin and ABA under drought stress, resulting in the mutant hypersensitive to drought and oxidation stresses with reduced photosynthesis capacity [182]. Two bZIPtype transcription factor genes, OsbZIP23 and Os-bZIP46 (also known as ABL1; see below), were characterized as positive regulators of stress tolerance in an ABA-dependent manner [183,184]. ABI5-Like1 (ABL1 or OsbZIP46), functionally analogous to its homolog in Arabidopsis ABI5 (ABA-insensitive5), is not only an important regulator of the ABA response, but also involved in auxin signaling, associated with the altered expression of several genes related to auxin metabolism or signaling [185].
The cellular redox homeostasis has profound effects in various physiological and pathological activities, including the stress response. The expression of OsTRXh1, encoding an H-type thioredoxin that acts to reduce thiol-disulfide bonds in proteins, is induced by salt and ABA. Transgenic studies showed that OsTRXh1 is involved in regulating the ABA response and tolerance to salt through modulating the ROS level [186]. Mutations in NITRIC OXIDE EX-CESS1 (NOE1) cause the accumulation of excessive amount of nitric oxide (NO) and hydrogen peroxide, resulting in cell death in leaves. NOE1 encodes a catalase responsible for scavenging hydrogen peroxide, suggesting that the interaction between two important signaling molecules NO and hydrogen peroxide plays a critical role in regulating cell death [187].

STRUCTURE, EVOLUTION AND REGULATION OF THE RICE GENOME
With the completion of whole-genome sequence and assembly of the rice genome [1,2], more attention was given to the dissection of complex agronomic traits as well as the regulation, origin and evolution of the rice genomes. In recent years, these studies are benefited greatly from the powerful nextgeneration sequencing technologies.
The Asian cultivated rice (Oryza sativa L.) has two subspecies, indica and japonica. The origins and domestication processes of these cultivars remain controversial for decades. A comprehensive map of rice genome variations was constructed through genome sequencing of 1,083 cultivated indica and japonica varieties as well as 446 geographically diverse accessions of the wild rice species Oryza rufipogon that is believed to be the immediate ancestral progenitor of cultivated rice. Analysis of these data revealed that the japonica cultivar was domesticated from O. rufipogon around the Pearl River in southern China, whereas the indica cultivar was originated from crosses between japonica and local wild rice as the initial cultivars. Several domesticationassociated traits are further analyzed through highresolution genetic mapping. This study not only clarifies a long-standing debt on the origin and domestication of the cultivated rice, but also provides important resources for rice breeding [188]. In a study of the evolutionary history of the rice genome by comparative analysis of 66 accessions from indica varieties, japonica varieties and their ancestral progenitor O. rufipogon, it was found that the genealogical histories of the overlapping low diversity regions are distinct from the genomic background. Moreover, while several known domestication genes were "rediscovered" by this approach, 13 additional candidate genes of domestication were identified [189].
The genus Oryza consists of 24 species, of which the Asian cultivated rice (O. sativa) is defined as A genome type, whereas the wild rice O. brachyantha is defined as F genome type located on the basal lineage in Oryza. The 261-Mb O. brachyantha genome was depicted by de novo sequencing and assembly. The O. brachyantha genome is characterized by low activity of long-terminal repeat retrotransposons and massive internal deletions of ancient long-terminal repeat elements, thus leading to a compact genome. Notably, more than 30% of the annotated genes of O. brachyantha or O. sativa are located in non-collinear position, presumably created mainly by double-strand break repair through non-homologous end joining. These findings provide important resources for functional and evolutionary studies in the genus Oryza [190].
By sequencing 517 rice landraces, more than 3.6 million SNPs were identified and a high-density haplotype map of the rice genome was constructed. Subsequent genome-wide association study (GWAS) on 14 agronomic traits in indica subspecies identified more than one third of the phenotypic variance [191]. In an independent study, resequencing the genomes of 40 cultivated accessions of rice and 10 accessions of their wild progenitors (O. rufipogon and O. nivara) identified 6.5 million high-quality SNPs, from which thousands of genes were characterized with significantly lower diversity in cultivated but not in wild rice, suggesting that these regions are extensively selected during domestication [192]. In another immense effort, a larger and more diverse sample of 950 worldwide rice varieties were analyzed and subjected to GWAS, resulting in the identification of 32 new loci associated with flowering time and 10 loci with grain-related traits [193].
Elite crop varieties usually fix alleles that show high diversity in non-elite gene pools through artificial selection. On the basis of this hypothesis, elite variety tag SNP alleles (ETASs) were identified by deep-sequencing six elite rice varieties and then comparing with two large control panels derived from published data [191,192]. Subsequent analysis of ETAS in the NCED gene, encoding a ratelimiting enzyme in the ABA biosynthetic pathway, revealed that this allele varied drastically in upland and irrigated rice varieties and is flanked by a selective sweep. Consistently, this allele is tightly associated with a higher level of ABA and denser lateral roots in the upland varieties, presumably due to artificial selection during breeding [194]. An additional example of artificial selection during domestication comes from a case study on the phenol reaction, which shows distinctive patterns in grains of indica and japonica cultivars. When treated with phenol solution, grains of indica, but not of japonica cultivars, turn brown, and the gene PHR1 responsive for this phenotype was characterized to encode a polyphenol oxidase. PHR1 is nullified by various mutations in all japonica lines but remains functional in nearly all indica and wild accessions, indicating of the important role of artificial selection driving the differentiation among domesticated varieties [195].
Transcriptome profiling at the whole-genome level is an important basis for functional genomics. The transcriptional activity of chromosome 4 in various tissues or organs at different developmental stages was analyzed by using a tilling microarray [196]. In a subsequent study by RNA-seq, the transcriptomes of two cultivated varieties indica and japonica were comprehensively analyzed and compared at single-nucleotide resolution [197]. Moreover, a genome-wide gene expression atlas was presented by analysis of 39 tissues collected throughout the life cycle in two indica varieties [198]. Together, these efforts have illustrated overall transcriptional landscapes of the rice genome and provided valuable resources to the international rice research community.
In the studies of epigenetic regulation of the rice genome, the biogenesis and regulation of small RNAs have gained considerable attention. The rice genome contains four ARGONAUTE1 (AGO1) genes. Four AGO1-related protein complexes were characterized, and the bound microRNAs (miR-NAs) and their targets were analyzed at the wholegenome level [199]. A systematic analysis defined the distinctive roles of OsDCL4, OsDCL3b and Os-DCL1 in the biogenesis of the 21-and 24-nucleotide (nt) phased small RNAs [200]. The characterization of a mutant in the RNA-dependent RNA polymerase 6 (OsRDR6) gene uncovered its critical role in small RNA biogenesis, spikelet development and antiviral defense [201,202]. A tissue-specific profiling analysis revealed the composition and expression pattern of pollen-specific or -enriched miRNAs, implying an important role of the miRNA pathway in pollen development [203]. Interestingly, a number of endogenous miRNA target mimics were identified, and the target mimics for miR160 and miR166 were demonstrated to be involved in the regulation of rice development [204]. MiRNAs in plants are usually 21 nt in length and bound to AGO1 protein to cleave specific target mRNA. A class of 24-nt long miRNAs was identified, which direct DNA methylation of their specific target genes. This finding defines a novel mechanism of miRNA in regulating gene expression [205].
The dynamic methylation of histones is a major means of epigenetic regulation. Two histone methyltransferases SDG724 and SDG725, catalyzing methylation at H3K36, were functionally characterized, which are involved in the regulation of flowering and BR signaling, respectively [206,207]. In a reverse reaction, the histone H3K4 demethylase JMJ703 was found to play a critical role in the regulation of transposon activity [208]. Remarkably, the analysis of an epi-allele of FIE1 links DNA methylation at the FIE1 locus with altered histone methylation patterns [209]. Together with a previous study on SDG714, which catalyzes histone H3K9 methylation to mediate DNA methylation [210], these findings revealed a novel regulatory mechanism that links histone methylation and DNA methylation, two important epigenetic marks.

PERSPECTIVES
During the past several years, Chinese scientists have made tremendous progresses for dissecting the complex agronomic traits in rice. As a part of these efforts, more than 140 agronomically important genes were identified and functionally characterized, which are involved in the regulation of almost every aspect of rice growth and development or the responses to stresses (see Table 1). In particular, breakthrough discoveries have been made in several important fields, including the elucidation of the regulatory mechanisms of shoot architecture, reproductive development, seed development, genome evolution and domestication. These progresses should allow us to dissect the related traits in more details, establish frameworks on these traits, and eventually utilize in molecular breeding. A complex trait is usually controlled by multiple loci and multiple pathways, and a single locus, in turn, is also involved in regulating, directly or indirectly, distinctive traits in many cases. The latter notion is perhaps the best evident by the observation that MOC1, IPA1, DEP1 and Ghd7 play important roles in regulating tillering, plant height and panicle development [3,19,74,90]. In rare cases, a gene is involved in the regulation of seemingly unrelated or separated biological processes, such as in XA13 and DST [95,96,167]. These findings illustrate the difficulties of integrating many important loci into a REVIEW Zuo and Li 271 network for full understanding the genetic and biochemical basis of the complex agronomic traits. Nevertheless, this task should be a major challenge for future studies, which will be continually based on the systematic analysis of agronomically important loci, especially the QTL-controlled traits. We noticed that many of the cloned genes encode novel proteins with unknown biochemical functions, thus preventing mechanistic understanding on the related traits. It also should be noted that many agronomic traits, especially those involved in adaptability in response to environmental alterations, are subjected to epigenetic regulation, and the underlying mechanisms are, unfortunately, poorly understood. Importantly, as demonstrated by several recent studies [135,189,191,193,194], the analysis of the dynamic changes of the rice genome during evolution and domestication in combined with GWAS, powered by the next-generation sequencing and other technologies, should allow fast and efficient identification of important trait-associated loci and alleles that are difficult to be identified by the genetic approach. With the molecular marker-assisted selection and newly developed genome editing technology [211,212], these agronomically important traits should be molecularly designed and eventually pyramided in the new generations of super rice varieties, as shown in the N411 accession that carries at least five positive grain-size known loci (qgl3, gs3, gw2, gw5 and GS5), thus producing extra-large grains [144].