Research Article

Cytokinin Oxidase Regulates Rice Grain Production

See allHide authors and affiliations

Science  29 Jul 2005:
Vol. 309, Issue 5735, pp. 741-745
DOI: 10.1126/science.1113373


Most agriculturally important traits are regulated by genes known as quantitative trait loci (QTLs) derived from natural allelic variations. We here show that a QTL that increases grain productivity in rice, Gn1a, is a gene for cytokinin oxidase/dehydrogenase (OsCKX2), an enzyme that degrades the phytohormone cytokinin. Reduced expression of OsCKX2 causes cytokinin accumulation in inflorescence meristems and increases the number of reproductive organs, resulting in enhanced grain yield. QTL pyramiding to combine loci for grain number and plant height in the same genetic background generated lines exhibiting both beneficial traits. These results provide a strategy for tailormade crop improvement.

Food shortage is one of the most serious global problems in this century. The United Nations Food and Agricultural Organization (FAO) estimates that 852 million people worldwide were undernourished in 2000 to 2002 (1). The global population, now at 6.4 billion, is still growing rapidly and is projected to reach 8.9 billion people by 2050 (2). Cereals are an important source of calories for humans, both by direct intake and as the main feed for livestock. About 50% of the calories consumed by the world population originate from three cereals: rice (23%), wheat (17%), and maize (10%) (3). To meet the expanding food demands of the rapidly growing world population, crop grain production will need to increase by 50% by 2025 (4).

Many agronomically important traits, including yield, are expressed in continuous phenotypic variation. These complex traits usually are governed by a number of genes known as quantitative trait loci (QTLs) derived from natural variations (5). QTL analysis has been employed as a powerful approach to discover agronomically useful genes (6-13).

Rice (Oryza sativa L.) is a staple food and has been established as a model monocot because it has the smallest genome size (390 Mb) among the major cereals (14), because its genome is syntenic with the genomes of other cereals (15), and because rice can be transformed easily. As a result, many molecular markers for rice have been developed, many mutants have been generated and stocked, and the complete genome of rice has been mapped and sequenced (14, 16-21). These accomplishments have greatly facilitated QTL analysis in rice. Grain number and plant height are important traits that directly contribute to grain productivity. Dwarf rice and wheat varieties were developed by classical plant breeding methods, contributing to the green revolution in the 1960s. Higher yields were obtained from these dwarf crops because their short stature reduced lodging, which is an agronomic term for bending of plants toward the ground after wind or rain storms (22-25). During the past decade, many attempts have been made to characterize QTLs for grain production and plant height; however, the genes involved in these QTLs have not been identified yet, and their chromosomal positions remain obscure. We aimed to identify genes of QTLs for grain number and plant height, not only to elucidate molecular mechanisms that regulate grain productivity but also to use these genes for breeding.

QTL analysis. A choice of parental lines that show wide phenotypic variation in the targeted traits is necessary for QTL analysis, because QTL detection is based on natural allelic differences between parental lines. We chose an indica rice variety, Habataki, and a japonica variety, Koshihikari, because they not only exhibit large variations in agronomically important traits but also have many molecular markers available (21). On average, Habataki plants are shorter than individuals of Koshihikari but produce more grains in their main panicle (Fig. 1, A to D).

Fig. 1.

QTL analysis and molecular cloning. (A) Gross morphology of Koshihikari and Habataki at maturity. Scale bar, 1 m. (B) Comparison of plant height at the heading stage in Koshihikari (Ko) and Habataki (Ha). (C) Panicle structure of Koshihikari, Habataki, and 5150. Scale bar, 20 cm. (D) Comparison of grain number in the main panicle of Koshihikari (Ko), Habataki (Ha), and 5150. Values in (B) and (D) are means with SD (n = 10 plants). (E) QTL map of grain number (Gn) and plant height (Ph) on rice chromosomes. (F) Location of Gn1 and Ph1 on chromosome 1. (G) Coarse linkage map and physical map of Gn1. (H) High-resolution linkage map of Gn1a. The number of recombinants between the molecular markers is indicated below the high-resolution map. (I) OsCKX2 structure and mutation sites in Habataki (blue) and 5150 (red). S indicates the site of amino acid substitutions. (J) Comparison of grain number per main panicle in Koshihikari (Ko-gn1a/gn1a), NIL-Gn1a/gn1a, and NIL-Gn1a/Gn1a. (K) Comparison of grain number per main panicle in nontransgenic and transgenic lines. 2 copy CKX2, transgenic TC65 plant carrying two copies of OsCKX2 derived from Koshihikari; TC65, control japonica line; CKX2/AS, transgenic rice carrying antisense OsCKX2 cDNA from Koshihikari. Values in (J) and (K) are means with SD (n = 10 plants).

We developed primary-mapping populations of 96 backcross inbred lines (BILs) derived from the cross between Habataki and Koshihikari. Both grain number and plant height seemed to be regulated by QTLs, as these traits were approximately normally distributed in the mapping population (fig. S1). QTL analysis detected five QTLs for increasing grain number (Gn) and four QTLs for plant height (Ph) (Table 1 and Fig. 1E). The most effective QTL for plant height, Ph1, was located close to the semi-dwarf 1 gene (sd1) that encodes gibberellin 20 oxidase (23-25). Comparison of SD1 between Habataki and Koshihikari revealed that Habataki had a 383-base pair (bp) deletion in the coding region of gibberellin 20 oxidase. The resulting loss of function caused the reduced plant height in Habataki. The deletion in the gibberellin 20 oxidase is the same as the causal variation found in IR8, a variety that helped lead to the green revolution in rice (23-25).

Table 1.

Putative QTLs for grain number (Gn) and plant height (Ph). QTL names are designated with the abbreviation of the trait name. NML, nearest marker locus of putative QTLs; PVE, phenotypic variation explained by each QTL; LOD, logarithm of odds.

QTL name Chromosome number NML Position of NML (cM) Change in effectView inline PVE LOD
Grain number
Gn1 1 BB-85 22.6 H92 44% 9.863
Gn2 4 AE-19 102.1 K41 10% 1.922
Gn3 10 AJ-65 30.2 H39 7% 1.38
Gn4 12 BI-20 30 K35 9% 1.701
Gn5 12 BB-23 47 K35 11% 2.147
Plant height
Ph1 1 BC-55 146.4 K24 30% 6.531
Ph2 5 BF-37 66.5 H10 7% 1.389
Ph3 6 CC-84 107.3 H11 9% 1.629
Ph4 12 BB-48 91.4 H9 8% 1.565
  • View inline* H indicates a Habataki-enhanced trait; K, a Koshihikari-enhanced trait. In grain numbers or cm of plant height

  • The most effective QTL for increasing grain number, Gn1 on chromosome 1, was selected for further analysis. The Habataki Gn1 allele is expected to produce ∼92 more grains per main panicle than the Koshihikari allele; Gn1 explains 44% of the difference in grain number between Habataki and Koshihikari (Table 1). So far, several QTLs associated with yield have been reported in rice. Some of these QTLs are located near the Gn1 region on the short arm of chromosome 1, suggesting they might be the same QTL (21). Although these QTL genes have not been identified and characterized yet, it is possible that the Gn1 locus contributes to increased grain productivity in various rice varieties. The importance of Gn1 for enhancing grain number in rice suggested that this QTL would be a good candidate for cloning.

    QTL cloning. In QTL cloning, producing nearly isogenic lines (NILs) carrying only one target QTL is necessary to eliminate the effects of other QTLs (5). Consequently, the QTL of interest in the NIL can be considered as a single Mendelian factor (26). We produced the NIL-Gn1 carrying the Gn1 region from Habataki in the Koshihikari background and used it for Gn1 mapping.

    We used 96 F2 individuals derived from heterozygote (Gn1/gn1) plants of NIL-Gn1 for coarse mapping of Gn1. We found that Gn1 consisted of two loci, QTL-Gn1a and QTL-Gn1b. QTL-Gn1a was mapped within 2 cM between the molecular markers R3192 and C12072S, whereas QTL-Gn1b mapped to the upper region of QTL-Gn1a (Fig. 1, F and G). Gn1a was chosen as the target for positional cloning, because the effects of Habataki Gn1a and Gn1b loci were almost identical and because the position of Gn1a between the two markers had been unambiguously determined. The Gn1a allele of Habataki was semidominant, because the grain number of heterozygote plants (Gn1a/gn1a) was intermediate between those of homozygote plants, gn1a/gn1a and Gn1a/Gn1a (Fig. 1J).

    About 13,000 F2 plants derived from heterozygotes (Gn1a/gn1a) of NIL-Gn1a were used for high resolution mapping of Gn1a. The candidate region of Gn1a was narrowed down to the 6.3 kb between the markers 3A28 and 3A20 (Fig. 1H). In this region, the Rice Genome Automated Annotation System (27) predicted one reading frame with high similarity to cytokinin oxidase/dehydrogenase (CKX), OsCKX2 (28) (Fig. 1H). The OsCKX2 of Koshihikari and Habataki consist of four exons and three introns and encode proteins of 565 or 563 amino acids, respectively. Comparison of the DNA sequences between the cultivars revealed several nucleotide changes, including a 16-bp deletion in the 5′-untranslated region, a 6-bp deletion in the first exon, and three nucleotide changes resulting in amino acid variation in the first and fourth exons of the Habataki allele (Fig. 1I).

    We also analyzed the nucleotide sequences of OsCKX2 in three alleles of high-yielding rice varieties from China, 5030, 5150, and 90B2. An 11-bp deletion in the coding region was detected in 5150, which produced more than 400 grains in the main panicle in our experimental field (Fig. 1, C and D). This deletion created a premature stop codon, suggesting that 5150 is null for OsCKX2 (Fig. 1I). The other two varieties had sequences identical to the Habataki allele. The coincidence of the OsCKX2 null allele and a higher grain number suggested that a reduction or loss of function of OsCKX2 enhanced grain production.

    To confirm that OsCKX2 corresponds to Gn1a, we produced transgenic plants expressing different levels of OsCKX2 and examined their grain yield. As Koshihikari and Habataki fail to regenerate shoots from the callus, we used the easily regenerable cultivar Taichung 65 (TC65), which possesses the Koshihikari allele of OsCKX2. Transgenic plants carrying two copies of the sense strand of OsCKX2 that was highly expressed showed reduced grain numbers compared to TC65. However, transgenic plants with antisense strands of OsCKX2 that had reduced levels of expression developed higher grain numbers (Fig. 1K and fig. S2). We conclude that the QTL for increased grain number, Gn1a, is OsCKX2.

    Molecular analysis of OsCKX2. Cytokinin (CK) was first discovered as a plant hormone that promotes cell division (29). It is now known to influence various aspects of plant growth and development, including seed germination, apical dominance, leaf expansion, reproductive development, and delay of senescence (30). Natural CKs such as trans-zeatin (tZ) and isopentenyladenine (iP) are N6-substituted adenine derivatives that generally contain an isoprenoid side chain (31). CKX preferentially and irreversibly degrades nucleobase CKs by cleavage of the unsaturated N6-isoprenoid side chains (31). This catabolic enzyme probably plays the principal role in controlling CK levels in plant tissues (32-34).

    To examine whether the OsCKX2 locus affects CK metabolism, we analyzed the levels of CKs in inflorescence meristems of Koshihikari, Habataki, NIL-Gn1a, and 5150. Although the contents of active tZ were similar in these lines, CK nucleotides (i.e., tZRMP and iPRMP) were substantially more abundant in Habataki, NIL-Gn1a, and 5150 than in Koshihikari (Fig. 2A). Because the CK metabolism modifying the adenine moiety is partially shared with the purine salvage pathway, nucleobase CKs are readily converted to the corresponding nucleotides and nucleosides (31). This metabolic flow plays an important role in the homeostasis of active CKs (31). In this context, the accumulation of the nucleotide- and nucleoside-species is explainable by the reduction in CKX activity in Habataki, NIL-Gn1a, and 5150, and the increased production of CK conjugates to reduce the overall CK activity (see below).

    Fig. 2.

    Molecular characterization of OsCKX2. (A) Comparison of CK levels in the inflorescence meristem of Koshihikari, Habataki, NIL-Gn1a, and 5150. tZ, trans-zeatin; tZR, tZ riboside; tZRMP, tZR 5′-monophosphate; iP, isopentenyladenine; iPR, iP riboside; iPRMP, iPR 5′-monophosphate; gFW, grams fresh weight. Values are means with SD (n = 3 measurements). (B) Enzymatic CKX activity in yeast cells transformed with empty vector (Vec), the OsCKX2 allele from Koshihikari (Ko), and that from Habataki (Ha). (C) Expression analysis by RT-Southern blot of OsCKX2 in various organs [leaf, root, shoot apex meristem (sam), culm, inflorescent meristem (ifm), flower (flw), and embryo (emb)] of rice. (D) OsCKX2 expression in the inflorescence meristem of Koshihikari (Ko), Habataki (Ha), NIL-Gn1a (NIL), and 5150. Actin was used as a control in (C) and (D). (E to H) GUS expression under the control of the OsCKX2 promoter: (E) longitudinal section of a culm (ifm, inflorescent meristem; n, node; in, internode; r, root); (F) young flower; (G) longitudinal section of node and internode; and (H) transverse section of internode. (I) Phylogenetic relationship of CKX proteins in rice and Arabidopsis. OsCKXs, O. sativa CKXs (table S1); AtCKXs, Arabidopsis CKXs (28). (J) Expression analysis of OsCKX1 to OsCKX11 in the inflorescence meristem of Koshihikari (Ko), Habataki (Ha), NIL-Gn1a (NIL), and 5150 by RT-Southern blot. The Koshihikari genomic DNA was used as a template for the positive control (KoG) in the polymerase chain reaction (PCR) with primers designed for each OsCKX. The PCR produced bands at higher molecular weights than those generated from cDNA because they contained intron sequences. The expected signal sizes of the cDNAs are indicated by arrowheads.

    To test whether OsCKX2 encodes an active enzyme in Koshihikari and Habataki, we isolated the cDNAs and expressed the proteins in budding yeast, Saccharomyces cerevisiae. Although several amino acids varied between the two OsCKX2 proteins, they catalyzed the cleavage of iP side chains with similar specific activities (Fig. 2B). This result shows that both alleles of OsCKX2 encode functional enzymes.

    We next studied the expression profiles of OsCKX2 in various organs by reverse transcription (RT)-Southern blotting, because RNA gel blot analysis did not detect any signals because of the low expression levels. The RT-Southern blot showed that OsCKX2 was preferentially expressed in leaves, culms, inflorescence meristems, and flowers (Fig. 2C). The highest levels of OsCKX2 expression in inflorescence meristems were found in Koshihikari. Transcript accumulation was less abundant in Habataki and NIL-Gn1a and extremely low in 5150 (Fig. 2D). As these differences indicated a correlation between OsCKX2 expression levels and grain number, they suggested that the phenotypic differences observed might have been caused by differential transcription of OsCKX2.

    We next examined the tissue specificity of OsCKX2 expression in transgenic rice harboring an OsCKX2 promoter::β-glucuronidase (GUS) construct. GUS expression was observed mainly in the vascular tissue in developing culms, inflorescence meristems, and young flowers in the T2 generation of transgenic plants (Fig. 2, E to H). The expression of OsCKX2 in inflorescence meristems might regulate the CK level to control flower number. CK is known to be translocated acropetally via the xylem and systemically via the phloem (35). The high levels of expression in these tissues suggest that OsCKX2 plays a role in regulating CK levels in the vascular system of developing culms, where CK is transported to the inflorescence meristems.

    At least 11 putative CKX genes (OsCKX1 to OsCKX11) are present in the rice genome (Fig. 2I and table S1). This redundancy suggests that the OsCKXs could be functionally differentiated by their temporal and spatial expression patterns. In tobacco and Arabidopsis, overexpression of CKX results in reduced levels of endogenous CK and lower meristem activity (33, 34). In transgenic Arabidopsis, overexpression of AtCKX3, the allele showing the highest similarity to OsCKX2 of the seven AtCKXs (Fig. 2I), reduced flower number because of a decreased rate of primordia formation in the flower meristem (34). Our findings are in agreement with these results from Arabidopsis.

    We examined the expression of all OsCKX genes in the inflorescence meristems of the four lines, Koshihikari, Habataki, NIL-Gn1a, and 5150, to elucidate why their phenotypes were different despite the apparent high redundancy of OsCKX genes in the rice genome. OsCKX2 was the dominant OsCKX expressed in Koshihikari inflorescence meristems (Fig. 2J), underlining its role in crop productivity. In contrast to OsCKX2, the expression patterns of all other OsCKX genes did not differ among these cultivars (Fig. 2J), indicating that OsCKX2 functions in inflorescence development.

    QTL pyramiding. In this experiment, Gn1a was identified as OsCKX2, a gene that increased grain number by ∼21% (Fig. 1J). Gn1b, another QTL, has not been identified yet and will be the next target for characterization. The Habataki alleles of Gn1a and Gn1b additively increase grain number; both components of Gn1 (Gn1a and Gn1b) are ideally suited for application to a practical breeding program. The NIL-Gn1 carrying the Gn1 locus (Gn1a and Gn1b) of Habataki has a heavier panicle weight and is more susceptible to lodging.

    To resolve the problem, we employed a QTL pyramiding breeding strategy. In this approach, desirable QTLs are combined through crossing of NIL-QTLs into a common genetic background. First, we developed an NIL carrying the Habataki sd1 allele in the Koshihikari background (Fig. 3A). This NIL-sd1 was ∼20% shorter than Koshihikari, as expected because of the effect of the sd1 allele (Fig. 3, A and B). Simultaneously, an NIL-Gn1 carrying the Habataki Gn1a+Gn1b chromosome fragment that produced ∼45% more grain than Koshihikari (Fig. 3, A and C) was also selected. The degrees of increase in grain number (45%) and reduction of plant height (20%) in the NILs corresponded to the phenotypic variation effects of Gn1 (44%) or Ph1 (30%), respectively, as predicted by the QTL analysis (Table 1). NIL-sd1+Gn1 was generated by crossing NIL-Gn1 and NIL-sd1 (Fig. 3A). The grain number in the main panicle was 26% higher, and plants were 18% shorter in this line than in Koshihikari (Fig. 3, A to C). The reduction in grain number for NIL-sd1+Gn1 (207 grains), as compared to NIL-Gn1 (237 grains), seemed due to pleiotropic effects of the sd1 allele. The same degree of grain number reduction was also found in NIL-sd1, relative to Koshihikari (Fig. 3C). The positive effect of Gn1, however, outweighed the negative effect of sd1, in NIL-sd1+Gn1 (Fig. 3C). Because total grain number per plant is the most important factor for increasing grain yield under field production conditions, grain numbers per plant were compared rather than grain numbers per main panicle. When the comparison was based on grain numbers per plant, 34% and 23% increases were found in NIL-Gn1 and NIL-sd1+Gn1, respectively (Fig. 3D). No difference in grain size was observed for these two NILs. Thus, Gn1 is a useful locus for increasing grain productivity.

    Fig. 3.

    Phenotypic characterization of NIL-QTLs. (A) Plant morphologies and chromosome maps of Koshihikari, NIL-sd1, NIL-Gn1, and NIL-sd1+Gn1. White and red scale bars indicate 1 m and 20 cm, respectively. (B) Comparison of plant height, (C) grain number in the main panicle, and (D) grain number in whole plants for Koshihikari, NIL-sd1, NIL-Gn1, and NIL-sd1+Gn1. Values in (B) to (D) are means with SD (n = 10 plants).

    Toward the application of QTLs for a new green revolution. We succeeded in cloning a QTL (Gn1a) that increased grain number in rice. Gn1a encodes OsCKX2, an enzyme that degrades bioactive CK. A null allele of OsCKX2 had been selected for increasing crop yield in a conventional breeding program in China. Genome synteny allows breeders to integrate traits and genes among cereals (15). For example, rice chromosome 1 shows regions of sequence similarity with chromosomes 3, 6, and 8 in maize (36), where some QTLs for grain yield traits have been mapped (37-40). Gn1a in rice might correspond to these maize QTLs, and orthologous CKX genes in other species might regulate yield in other cereal crops as in rice.

    In molecular studies of rice and wheat varieties, the phytohormone gibberellin has been identified as a key player in controlling crop plant architecture (22-25). We demonstrate here that CK metabolism also contributes to crop productivity. Because CK controls cell division and lateral meristem activity (30, 31), CK accumulation in the inflorescence meristem can explain the significantly higher grain numbers.

    Identification of agronomically important QTLs and pyramiding of such QTLs presents a useful strategy for efficient crop breeding. Interspecific crosses between O. sativa and wild relatives could lead to the discovery of useful QTLs from a range of allelic variations much wider than those present in cultivated lines. Furthermore, wild rice species are likely to provide access to QTLs not only for yield but also for disease resistance, stress tolerance, and other desirable traits (6-8, 21), because these plants have adapted to unique geographic and environmental conditions. Discovering useful genes, improving agricultural traits hidden in the plant genome, and applying these findings to crop breeding will pave the way for a new green revolution.

    Supporting Online Material

    Materials and Methods

    Figs. S1 and S2

    Table S1

    References and Notes

    References and Notes

    Stay Connected to Science

    Navigate This Article