Functional Characterization of Genes/QTLs for Increasing Rice Yield Potential

Rice (Oryza sativa L.) as the monocot model plant and an important food crop is cultivated worldwide. Due to the rapid growth of the world’s population, rice yield is urgently required to increase to meet world food demand. In the last century, rice yield experienced rapid growth twice in China, which is mainly attributed to the exploitations of semi-dwarf 1(sd1) gene and heterosis of F1 hybrid. Before the green revolution, rice varieties were tall and had a low harvest index. Introgression of sd1 into the varieties significantly reduced the plant height and increased the harvest index, which resulted in a dramatic increase of rice productivity [1]. Heterosis breeding has been widely used to improve rice yield potential. Hybrid rice varieties usually have a yield advantage of 10-20% over the conventional inbred varieties, thus cover more than half of the total rice area in China at present [2, 3]. However, rice yield per unit area has not been much elevated and the arable land for rice cultivation has kept decreasing during the past two decades. New genetic improvement strategies are urgently required to break the bottleneck of yield potential of current varieties, which largely rely on the elucidation and exploitation of genetic and molecular basis for rice yield traits [4].


Introduction
Rice (Oryza sativa L.) as the monocot model plant and an important food crop is cultivated worldwide. Due to the rapid growth of the world's population, rice yield is urgently required to increase to meet world food demand. In the last century, rice yield experienced rapid growth twice in China, which is mainly attributed to the exploitations of semi-dwarf 1(sd1) gene and heterosis of F 1 hybrid. Before the green revolution, rice varieties were tall and had a low harvest index. Introgression of sd1 into the varieties significantly reduced the plant height and increased the harvest index, which resulted in a dramatic increase of rice productivity [1]. Heterosis breeding has been widely used to improve rice yield potential. Hybrid rice varieties usually have a yield advantage of 10-20% over the conventional inbred varieties, thus cover more than half of the total rice area in China at present [2,3]. However, rice yield per unit area has not been much elevated and the arable land for rice cultivation has kept decreasing during the past two decades. New genetic improvement strategies are urgently required to break the bottleneck of yield potential of current varieties, which largely rely on the elucidation and exploitation of genetic and molecular basis for rice yield traits [4].
Rice yield traits are complex agronomic traits governed by multiple genes called as quantitative traits loci (QTLs), which usually show a continuous phenotypic distribution in a segregating population derived from a cross of a pair of inbred lines. Most QTLs for yield traits show small genetic effect and are difficult to be identified. These minor QTLs play a vital role in regulating yield trait and are widely utilized in commercial rice varieties, so that finemapping and map-based cloning of these QTLs will be beneficial for breeding. Number of panicles per plant, number of grains per panicle, and grain weight are three component traits which are determined by tiller, panicle and grain development. Dissecting the genetic basis of these traits by QTL mapping can facilitate breeding high yield varieties. However, it is rather difficult to isolate these QTLs because each contributes little effect to yield traits, and the effect is strongly affected by the environment. In recent years, tremendous progress has been attained and many QTLs for rice yield traits have been isolated and functionally analyzed in detail, which provides new sights into the molecular mechanisms of the formation of rice yield traits. Meanwhile, mutant analysis has also functionally characterized many genes involved in yield traits because of the availability of rice genome and rice mutant collections. These studies greatly strengthen our understanding of regulatory mechanisms of these traits. In this chapter, we summarize the recent progress in the genetic and molecular mechanisms underlying rice yield traits and illustrate a strategy to develop varieties with higher yield potential.

Identification of QTLs
QTL mapping of a target trait is defined as the chromosomal location and genetic characterization of QTLs for the trait through the association between genetic markers and phenotypic variations. To facilitate this mapping, development of mapping population, construction of linkage map and phenotypic evaluation are essential for QTL analysis.
Typically, mapping population includes F 2 plants, doubled haploid lines (DHLs) and recombinant inbred lines (RILs). F 2 population that carries the complete genetic information from the parents can be easily developed, but its phenotypic evaluation cannot be replicated [5]. Due to the inherent homozygosity in the lines, both DHL and RIL populations can be planted repeatedly in different planting seasons and environment conditions as many times as necessary. DHL population is limitedly used in QTL mapping due to the difficulties in the plant regeneration from cultured anthers, especially for indica rice varieties [6,7]. RIL population is widely used in QTL mapping although it is time-consuming and labor-intensive to prepare the population. Many RIL populations have been developed from inter-subspecies crosses [8,9], intra-subspecies crosses [10][11][12], or crosses between commercial cultivars and wild rice [13].
Linkage map is composed of many linkage groups according to different chromosomes, which are constructed by genotyping using genome-wide polymorphic markers. DNA based molecular markers, such as restriction fragment length polymorphism (RFLP), simple sequence repeat (SSR), cleaved amplified polymorphic sequence (CAPS) and single nucleotide polymorphism (SNP), are widely applied in the construction of linkage map. Based on the complete genome-wide sequence of rice, it becomes easier to design genome-wide polymorphic markers and construct high density molecular linkage map [14].
Yield trait conditioned by QTLs usually varies continuously in a mapping population. Phenotypic values are difficult to be accurately measured due to environmental influences, especially for F 2 population without replication. The precision of phenotypic data greatly affects the resolution of QTL mapping [15].
Thousands of QTLs for rice yield traits have been detected and are distributed throughout the whole genome while many of them are co-localized (http://www.gramene.org). We use one of our studies to illustrate the general process for the classical characteristics of QTLs ( Figure  1). A RIL mapping population is developed from an indica-indica cross between Zhenshan 97B and Milyang 46 using single seed descent method. QTL analysis shows that each yield trait is controlled by several QTLs. These QTLs are dispersedly distributed on chromosomes, and function on yield productivity not only by their own effects, but also by within-locus and interlocus interactions [11].  (Figure 1). Following MAS in a high density marker linkage map, a series of RHLs with overlapping segregated segments for the target QTL are selected from F 7 RILs. This method has proven to be efficient, and several yield trait QTLs, such as qGY6 and qGL7-2, have been successfully fine-mapped and validated [2,16].

Zhenshan
Introgression lines (ILs) and chromosome segment substitution lines (CSSLs), which are developed by backcrossing repeatedly with the recurrent parent, can also be used for QTL validation, fine-mapping and breeding superior rice varieties [17,18].

QTLs/genes for tillering
Rice tillers are mainly composed of primary, secondary and tertiary tillers, which are shoot branches arising from the unelongated basal internodes. Tillering starts with the appearance of the fourth leaf from the main culm. Usually, the duration of tillering will last about 25-30 days. The number of panicles and yield potential are determined by panicle-bearing tillers, and grain yield are mainly contributed by primary and secondary tillers. Therefore, tiller number is considered a key component in determining rice yield. Some key genetic factors responsible for rice tillering have been molecularly characterized ( Figure 2 and Table 1).
Rice tillering undergoes two major processes, the formation and outgrowth of tiller bud. Isolation and characterization of MONOCULM (MOC1) provide a new sight for the formation of tiller bud. The moc1 mutant phenotypically exhibits only one main culm without any tillers due to the deficiency to form axillary bud. MOC1 encodes a member of GAI, RGA and SCR (GRAS) family nuclear proteins to function on the formation of axillary buds [19].  Phytohormone pathways play a crucial role in controlling the outgrowth of tiller bud from leaf sheath. Plant hormones interact to regulate axillary bud outgrowth. It is well known that auxin maintains shoot apical dominance and inhibit axillary bud outgrowth, whereas cytokinins promote branches development [4]. Strigolactones, as a new kind of terpenoid plant hormones, might act as the downstream of auxin to inhibit axillary bud outgrowth. Several genes involved in the synthesis and signaling pathway of strigolactones are isolated and functionally characterized through analyzing a serious of tillering dwarf mutants [20,21]. DWARF17 (D17)/ HIGH-TILLERING DWARF1 (HTD1), DWARF10 (D10) and DWARF27 (D27) are involved in the biosynthesis of strgolactones, while DWARF3 (D3) and DWARF14 (D14) act in the signaling pathway in rice [22][23][24][25][26]. Their loss-of-function causes similar phenotype of enhanced shoot branches accompanying with reduced plant height. Although the relationship among phytohormones in regulating axillary bud outgrowth is complex and requires more proof to substantiate, recent advances in the regulatory mechanisms involved in phytohormones help further understand rice tillering.
Tiller number and angle are major determinants of rice plant architecture. New plant type known as ideal plant architecture (IPA) is proposed with reduced tiller number with almost no unproductive tillers to improve cultivar yield potential. A major QTL for IPA encoding SOUAMOSA PROMOTER BINDING PROTEIN-LIKE 14 (OsSPL14) has been cloned. OsSPL14 is regulated by microRNA OsmiR156, and increasing level of the OsSPL14 transcript and protein results in an IPA phenotype and higher grain productivity [27,28]. PROSTRATE GROWTH 1(PROG1) controlling wide tiller angle and great number of tillers in wild rice species encodes a zinc-finger nuclear transcription factor and is highly expressed in the axillary meristems. An amino acid substitution caused by a SNP in PROG1 leads to the transition from prostate growth of the wild rice O. rufipogon to erect growth of the domesticated rice O. sativa [29,30]. In addition, qGY2-1, a major QTL for grain yield per plant, encodes leucine-rich repeat receptor-like kinase (LRK), and over-express of LRK1 causes more tillers and greater grain yield than the wild type [31].

QTLs/genes regulating number of grains per panicle
Number of grains per panicle is an important agronomic trait for grain productivity, which is determined by the panicle formation. During the past two decades, many genes/QTLs controlling panicle development have been characterized ( Figure 2 and Table 1). Rice panicle developed from a terminal inflorescence at the top of a stem contains panicle axis, primary and secondary branches, pedicel and spikelets. Pedicels arise from the primary and secondary branches and bear spikelets on the top. Panicles and the bearing spikelets on them directly determine the rice yield.
Inflorescence development determines the formation of rice panicle. Inflorescence meristem generates primary and secondary branches meristems, and subsequent spikelet meristems. Several genes involved in the formation of inflorescence branch and spikelet meristems are identified through mutant analysis. LAX1 encodes a basic helix-loop-helix (bHLH) transcription factor and is required for the initiation/maintenance of inflorescence branch meristems. The lax1 mutant produces severely reduced primary and secondary branches and spikelets [32]. FRIZZY PANICLE (FZP), which encodes an ethylene-responsive element-binding factor (ERF), is responsible for the establishment of floral meristem identity through suppressing the formation of axillary meristems within the spikelet meristem. The fzp mutant is deficient in spikelet development and exhibits sequential rounds of branching instead of the formation of florets [33]. ABERRANT PANICLE ORGNIZATION (APO1), which encodes an F-box protein, functions in preventing the precocious transition from branch meristems to spikelet meristems. The apo1 loss-of-function mutants produce small panicles with greatly reduced branches and spikelets [34]. In addition, APO2 interacts with APO1 to regulate panicle development [35].
Rice panicle size is largely determined by the number and length of primary and secondary branches. SHORT PANICLE 1 (SP1) encodes a putative transporter of the peptide transporter family and participates in the elongation of rice panicle. The mutation of SP1 causes a shortpanicle phenotype due to the defect in the elongation of inflorescence branches in the sp1 mutant [36]. OsSPL14 not only controls tillering, but also promotes panicle branching and produces larger panicles with more spikelets [27,28].
Rice panicle architecture is mainly determined by the arrangement of primary and secondary branches and grain density. Erect panicle is an important agronomic trait closely related to grain yield. DENSE AND ERECT PANICLE1 (DEP1) encodes a phosphatidylethanolaminebinding (PEBP) protein-like domain protein and controls panicle branches, grain density and erect panicle. The gain-of-function mutation in DEP1 resulted in the phenotype of increased primary and secondary branches and number of grains per panicle, and decreased panicle length [37]. DEP2, which encodes a plant-specific protein and is strongly expressed in young panicles, is responsible for panicle outgrowth and elongation. The dep2 mutant displays a dense and erect panicle phenotype [38].
Cytokinins regulate number of spikelets per panicle. A major QTL, GRAIN NUMBER1 (Gn1a), which encodes cytokinin oxidase/dehydrogenase (OsCKX2), controls number of grains per panicle. Repression of OsCKX2 leads to cytokinin accumulation, which finally results in the increase of number of grains per panicle and grain yield [39]. LONELY GUY (LOG) is responsible for shoot meristem activity and encodes cytokinin-activating enzyme for the conversion from inactive cytokinin nucleotides to the free-base forms. Loss of function of LOG results in producing small panicles with reduced panicle branches and grains in the log mutant [40]. LARGER PANICLE (LP) encoding a Kelch repeat-containing F-box protein regulates panicle architecture. Larger panicle with more primary branches and grains is observed in the LP lossof-function mutants. LP could regulate panicle architecture by modulating cytokinin level due to the significant down-regulation of OsCKX2 expression level in the mutants [41]. Furthermore, DEP1 might control the number of panicle branches through cytokinin pathway because expression level of OsCKX2 is clearly down-regulated in NIL-dep1 plant [37]. These studies imply that the phytohormone cytokinin plays a vital role in regulating panicle development.

QTLs/genes controlling grain weight
Rice grain is closely enclosed by a hull which is composed of one palea, lemma, rachilla and two sterile lemmas. A brown rice mainly consists of bran, endosperm and embryo. During the process of grain filling, endosperm cells expand and accumulate a massive amount of nutrients, mainly starch. Rice grain weight is largely determined by the endosperm size. Dozens of genes/QTLs involved in rice grain weight have been isolated and molecularly characterized ( Figure 2 and Table 1).
Given that each grain in a rice panicle can be fully filled, grain weight is determined by grain size, which can be measured with grain length, width and thickness. GS3, GL3.1/qGL3 and TGW6, three major QTLs controlling grain length, are map-based cloned and functionally analyzed [42][43][44][45]. GS3 encodes a putative trans-membrane protein containing four putative domains, a plant-specific organ size regulation (OSR) domain, a trans-membrane domain, a tumor necrosis factor receptor/nerve growth factor receptor (TNFR/NGFR) family cysteinerich domain and a von Willebrand factor type C (VWFC). Loss-of-function or deletion of plantspecific OSR domain results in long grain phenotype [42]. GL3.1/qGL3 encodes Ser/Thr phosphatase of phosphatase kelch family to regulate grain length and yield. GL3.1/qGL3 directly down-regulates Cyclin-T1;3 to dephosphorylate Cyclin-T1;3 and results in short grain shape [43,44]. THOUSAND-GRAIN WEIGHT 6 (TGW6) controls grain length and weight, which expression is especially high around the endosperm in the pericarp. TGW6 possesses indole-3-acetic acid (IAA)-glucose hydrolase to decompose IAA-glucose into IAA and glucose, which influences the transition timing from the syncytial to the cellular phase and results in short grain phenotype [45]. SMALL AND ROUND SEED (SRS3), which encodes a protein of the kinesin 13 subfamily containing a kinesin motor domain and a coiled-coil structure, is strongly expressed in developing organs and regulates rice grain length. The srs3 mutant shows shorter cells compared to the wild type, which causes the small and round seed phenotype [46]. Srs5 encodes alpha-tubulin protein and its mutation produces a semidominant mutant exhibiting similar phenotype with the srs3 mutant [47]. POSITIVE REGU-LATOR OF GRAIN LENGTH 1(PGL 1) and ANTAGONIST OF PGL1 (APG) encode an antagonistic pair of bHLH proteins and interact to regulate rice grain length [48]. PGL 1 and PGL 2 redundantly suppress the function of APG to form elongated grains [49]. In addition, brassinosteroid (BR) pathway affects rice grain size. A series of mutants related to the synthesis and signaling pathway of BR such as d61, brd1and short grain1 (sg1) display shorter grain phenotype than their wild types [50][51][52].
Four QTLs conditioning grain width, GW2, qSW5/GW5, GS5 and GW8, have been isolated and characterized. GW2 encodes a RING-type protein with E3 ubiquitin ligase activity to function in the protein degradation through the ubiquitin-proteasome pathway. GW2 E3 ligase negatively regulates cell division and the mutant allele of GW2 promotes spikelet hull cell division to result in an increase of grain width and weight [53]. GW5 and qSW5 are the same QTL on chromosome 5 in fact, identified by two research groups separately [54,55]. qSW5/GW5 encodes a novel nuclear protein, physically interacting with polyubiquitin and acting in the ubiquitin-proteasome pathway to regulate cell division. qSW5/GW5 is also a negative regulator for grain width and its mutant allele causes an increase of grain width [54]. GS5 encodes a putative serine carboxypeptidase and positively regulates grain width. Over expression of GS5 promotes cell division and results in increased grain width [56]. GW8, synonym with OsSPL16, encodes a positive regulator of cell proliferation and conditioning grain width and yield. Enhanced expression level of GW8 promotes cell division and grain filling, while its lossof-function forms a slender grain [57].
Grain thickness largely depends on the ability of grain filling. GRAIN INCOMPLETE FILLING 1(GIF1), which encodes a cell-wall invertase to download sucrose in the ovular and stylar vascular tissues and hydrolyzes sucrose to glucose and fructose for the starch synthesis in the endosperm, is responsible for rice grain-filling and yield. Mutation in the GIF1 causes slower grain-filling to result in reduced levels of glucose, fructose and sucrose in the gif1 mutants. Compared to the wild type GIF1, the cultivated GIF1 displays a restricted expression during the filling stage to bring about grain weight increase [58]. Expression level of GIF1 is substantially low in the heading and grain weight (hgw) mutant, which delays the heading date and reduces grain weight. HGW encodes a novel plant-specific ubiquitin-associated domain protein and acts through GIF1 to regulate grain width and weight [59]. FLOURY ENDO-SPERM2 (FLO2) encodes a protein harboring a tetratricopeptide repeat motif and preferentially expresses in developing seeds. FLO2 positively regulates the expression of genes involved in production of storage starch and proteins in the endosperm, so mutation of FLO2 causes significantly smaller grain size phenotype [60].

QTLs/genes for rice yield-related traits
Plant height and heading date are two important agronomic traits closely related to rice yield. The Green Revolution has made a tremendous contribution to solve the global food crisis, and this mark achievement in rice is caused by the application of sd1 gene. SD1 encodes an oxidase enzyme involved in the biosynthesis pathway of gibberellin, which is one of the most important determining factors of plant height. Mutation of SD1 produces semi-dwarf phenotype and significantly increase rice yield [61].
Genes/QTLs controlling heading date usually prolong the duration of panicle differentiation to produce more spikelets per panicle and enhance grain yield potential.

Future perspectives
As mentioned above, cloning and functional characterization of genes/QTLs have greatly strengthened our understanding in the genetic and molecular mechanisms underlying rice yield traits, which has facilitated the breeding efforts for higher yield potential varieties. Pyramiding of favorable genes/QTLs has become an efficient strategy in rice genetic improvement and is widely adopted by rice breeders. For instance, combination of Gn1 (Gn1a+Gn1b) and sd1 into Koshihikari has simultaneously improved two traits with increased grain numbers per plant by 23% and reduced plant height by 18% as compared to wild type Koshihikari [39]. The NIL (GW8/gs3) with a pyramiding of GW8 and gs3 produces longer and wider grains than the wild type NIL (gw8/GS3) [57].
Although tremendous progress has been made in the studies of rice yield trait, there is still a long way to clearly elucidate the molecular mechanisms responsible for the formation of rice yield traits. Almost all the rice yield traits including number of panicles per plant, number of grains per panicle and grain weight exhibit comprehensive and continuous variations in the genetic population or among the commercial varieties, typically due to the function of multiple genes called as QTLs. According to the Gramene database, thousands of QTLs conditioning rice yield traits have been detected by QTL mapping and majorities of them are minor QTLs with small genetic effect, which are difficult to be identified through mutant analysis. However, minor QTLs may participate in different molecular pathways to regulate rice yield traits and play a vital role in improving yield potential. During the long domestication process, these minor QTLs have been selected and combined relying on the breeders' experience to develop cultivated varieties. Therefore, more efforts are necessary to isolate minor QTLs and elucidate the functional mechanisms in the future.
Natural variation exists widely in the genes/QTLs, resulting in many alleles for each gene/QTL. For example, Ghd7 has at least five alleles including Ghd7-0, Ghd7-0a, Ghd7-1, Ghd7-2 and Ghd7-3 which enable rice to be cultivated in different ecotype regions. Till now, it is still rather difficult to combine favorable alleles freely in breeding higher yield potential varieties. Mining the alleles is a key base to combine the alleles. Based on the affordable next-generation sequencing technology, association mapping is a promising strategy to mine favorable alleles using a set of diverse germplasm accessions. On the basis of available and favorable alleles, an efficient breeding strategy has been proposed to exploit rice yield potential, involved in identifying the genes/QTLs for rice yield traits, mining alleles of target genes/QTLs through candidate-gene association mapping, developing functional markers and combining favorable alleles in cultivated varieties ( Figure 3).