Genetic control of inflorescence architecture during rice domestication

Inflorescence architecture is a key agronomical factor determining grain yield, and thus has been a major target of cereal crop domestication. Transition from a spread panicle typical of ancestral wild rice (Oryza rufipogon Griff.) to the compact panicle of present cultivars (O. sativa L.) was a crucial event in rice domestication. Here we show that the spread panicle architecture of wild rice is controlled by a dominant gene, OsLG1, a previously reported SBP-domain transcription factor that controls rice ligule development. Association analysis indicates that a single-nucleotide polymorphism-6 in the OsLG1 regulatory region led to a compact panicle architecture in cultivars during rice domestication. We speculate that the cis-regulatory mutation can fine-tune the spatial expression of the target gene, and that selection of cis-regulatory mutations might be an efficient strategy for crop domestication.

I nflorescence architecture is a key agronomical factor determining grain yield, and thus has been a major target of crop domestication and improvement 1,2 . Understanding the genetic basis of inflorescence architecture will contribute to not only elucidating the crop evolutionary mechanism but also improving crop grain yield.
Asian cultivated rice (Oryza sativa L.), one of the oldest domesticated crops, now feeds over one-third of the global population 3 . Morphological and physiological traits had been remarkably changed during the domestication of rice. Recently, evolutionary mechanisms and genetic factors controlling rice domestication-related traits, such as grain shattering, prostrate growth habit, pericarp colour and grain size have been characterized [4][5][6][7][8][9][10][11][12] . The transition in inflorescence architecture from a spread panicle to a compact panicle was another critical event for rice domestication. Common wild rice (O. rufipogon Griff.) displays typical architecture of a spread rice panicle (Fig. 1a), which may be adapted to its out-crossing pollination habit, contribute to seed dispersal at maturity and reduce the risk of pathogen infection 13,14 . These characteristics of wild rice are beneficial to maintain a high genetic diversity and permit adaptive changes to natural ecological environments. In contrast, cultivated rice has a compact panicle (Fig. 1b), which may be more amenable to high-density planting once rice plants grow erect, improve harvest efficiency, bear more seeds on the panicle branches and enhance photosynthetic efficiency of the lower leaves 2,15 . These characteristics of cultivated rice may increase grain-yield potential and meet the human's needs of cultivation.
The spread panicle of O. rufipogon has been a classical trait for genetic studies of wild rice. In 1932, Mitra and Ganguli 16 were the first to report that the spread panicle of wild rice was controlled by two complementary dominant genes This was confirmed in 1990 when a dominant single locus controlling panicle spread in wild rice was identified on chromosome 4 (ref. 17). However, until now the molecular basis for this key event in rice domestication has not been elucidated.
Here we characterize an introgression line that exhibits spread panicle phenotype and find that the spread panicle architecture of wild rice is controlled by a dominant gene, OsLG1, a previously reported SBP-domain transcription factor that controls rice ligule development 18 . Association analysis indicates that a singlenucleotide polymorphism (SNP)-6, residing in the cis-regulatory region 11 kb upstream of the translation start site, affects the OsLG1 gene expression, which is associated with a significantly altered cell morphology at the panicle pulvinus, and consequently leads to a compact panicle architecture in domesticated rice cultivars.

Results
Characterization of the spread panicle introgression line YIL31. In an attempt to identify the gene responsible for a spread   19 . One introgression line (YIL31) displaying a spread panicle, which harboured an YJCWR chromosome segment on the long arm of chromosome 4 ( Supplementary  Fig. S1), was selected for further study. Morphological analysis revealed that the panicle branch angle of YIL31 was much greater than that of Teqing (Fig. 1c,d), and a panicle pulvinus was present in YIL31 but absent in Teqing (Fig. 1e,f). Histological examination showed that the panicle pulvinus in YIL31 was composed of several layers of expanded parenchyma cells, which prevented the panicle branch from growing erect and resulted in a more angled growth pattern ( Fig. 1g-j).
Map-based cloning of OsLG1. Preliminary genetic analysis of an F 2 population (285 progeny) derived from the cross between YIL31 and Teqing suggested that the spread panicle phenotype was controlled by a single dominant gene. Genetic linkage analysis with 1,368 F 2 progeny revealed that the functional mutation (FM) of the target gene was located between markers M1 and M2 on chromosome 4 (Fig. 2a). Then, we used a much larger F 2 population, consisting of 15,198 plants, and delimited the FM to a 3.3-kb region between the M5 and M6 markers (Fig. 2b). M5 and M6 were used to identify a single BAC clone (YJ0510607) from the YJCWR genomic BAC library (Fig. 2c). Sequence analysis indicated that the fragment contained noncoding DNA as no predicated gene sequence was found within the 3.3-kb region. Two subcloned segments V1 (11.6-kb) and V2 (13.7-kb), harbouring the 3.3-kb region, were used to transform japonica cultivar Zhonghua 17 (with a compact panicle). All 44 independently generated transgenic lines (19 V1 lines and 25 V2 lines) still showed the compact panicle phenotype similar to Zhonghua 17, indicating that complementation had not occurred.
Further sequence analysis of the surrounding region revealed that the OsLG1 gene, a previously reported SBP-domain transcription factor that controls rice ligule development 18 , is located 10 kb from the finely mapped region. To test whether OsLG1 is responsible for the spread panicle phenotype, we introduced a 28.5-kb genomic segment (V3) harbouring the entire 3.3-kb fine-mapping region and the OsLG1 into Zhonghua 17 (Fig. 2c). All nine independently generated transgenic lines showed full complementation with a spread panicle phenotype similar to that of YIL31 (Fig. 2d).
We sequenced the coding region of OsLG1 and found no sequence variation between YIL31 and Teqing. We then compared the sequences of the 3.3-kb fine-mapping region between YIL31 and Teqing, and identified 12 SNPs and one 6-bp insertion/deletion variations ( Supplementary Fig. S2). As the 3.3-kb region contained non-coding DNA, it is possible that these sequence variations in the 3.3-kb-mapping region may affect the OsLG1 expression contributing to the transition of panicle architecture. Consistent with this speculation, we found that the mRNA expression level of OsLG1 was much higher at the panicle pulvinus in the transgenic plants (spread panicle phenotype), compared with the control plant (compact panicle phenotype; Fig. 2e). Further confirmation was achieved by overexpressing the OsLG1-coding region, driven by the strong constitutive ubiquitin promoter, which induced the spread panicle phenotype in Zhonghua 17 transgenic lines (Fig. 2d,e). These results demonstrated that strong expression of OsLG1 led to spread panicle architecture.
In addition to YIL31, we identified another introgression line CIL49 derived from a cross between YJCWR and compact panicle japonica cultivar C418. The CIL49 displayed similar spread panicle as YIL31 (Fig. 2f,g), and also had a panicle pulvinus that caused the larger angle between the panicle branch and main stem (Fig. 2h,i). Genotypic analysis also showed that the YJCWR segment was present in the long arm of chromosome 4 in CIL49. Sequence analysis indicated that no variation existed in OsLG1, whereas the 12 SNPs were present in the 3.3-kb region between CIL49 and C418. Quantitative reverse transcriptase (RT)-PCR analysis showed that OsLG1 expression at the panicle pulvinus in CIL49 (spread panicle phenotype) was higher than that in C418 (compact panicle phenotype; Fig. 2j, Supplementary Table S2).
Taken together, these data suggest that variations in the 3.3-kbmapping region of cultivated rice, located 10-kb upstream from the OsLG1 translation start site, decreased the OsLG1 gene expression at the panicle pulvinus, consequently leading to the transition from spread to compact panicle.
An SNP caused a compact panicle in domesticated rice. To determine which mutation was associated with the phenotype, we sequenced the 3.3-kb fine-mapping region of 158 compact panicle cultivars from 17 countries and 21 spread panicle accessions of O. rufipogon (Supplementary Table S1). We found that the 21 accessions of wild rice clustered in 17 haplotypes (H1-H17), whereas the 158 cultivars clustered in only 2 haplotypes (H18 and H19; Fig. 3a). The haplotype H18 included Teqing and other seven cultivars, whereas the haplotype H19 included all the remaining 150 cultivars, suggesting that the OsLG1 allele was strongly selected in cultivated rice. An association test with panicle architecture phenotypes and the 12 SNPs and one 6-bp insertion/deletion identified in wild and cultivated rice revealed that the strongest signal was present at the SNP6 site (P ¼ 1.2 Â 10 À 26 ; Fig. 3b). Medium signals were detected at the SNP2, SNP4 and SNP12, which were in high linkage disequilibrium with the SNP6 (r 2 40.6; Fig. 3c). Further sequences comparisons demonstrated that the nucleotide at the SNP6 site of all the 21 wild rice accessions is G, whereas that of all the 158 cultivars is A (Fig. 3a). This nucleotide change, G to A, at the SNP6 site was completely consistent with the transition of panicle architecture. These results suggested that the SNP6 mutation in the regulatory region was responsible for compact panicle architecture of cultivated rice.
To test whether the 3.3-kb regulatory region had been a target for artificial selection during rice domestication, we analysed the signature of selection in 3.3 kb sequences from a panel of 19 wild rice (O. rufipogon) and 62 cultivated varieties. The nucleotide diversity was nearly zero in the cultivated rice, which is significantly lower than that of wild rice (Fig. 3d). This result indicated that the 3.3-kb regulatory region had experienced strongly artificial selection during rice domestication.
Phylogenetic analysis of OsLG1. The OsLG1 gene encodes the SBP-domain transcription factor OsSPL8, which encodes 416 amino acids (Fig. 4a). We retrieved through BLASTp searches using the full-length protein sequence of OsLG1 as a query against the non-redundant protein database (http://www.ncbi. nlm.nih.gov/BLAST/), and identified 14 putative homologues in other crop plants. Phylogenetic analysis of these 14 putative homologues plus 35 SBP-domain genes previously identified in rice and Arabidopsis indicate that OsLG1 is most closely related to genes found in other monocots, including sorghum (Sb06g31290), maize (ZmLG1) 20 , barley (HvLG1) and Brachypodium (BdSPL8; Fig. 4b). The SBP domains were highly conserved between OsLG1 and its homologues in other crop plants, noting that the predicted SBP domains of OsLG1, Sb06g31290 and ZmLG1 were completely identical (Fig. 4c).
Expression pattern of OsLG1. The OsLG1 gene belongs to the SPL transcription factor family related to the plant architecture and inflorescence development [21][22][23][24][25] . When we introduced the OsLG1-GFP (green fluorescent protein) fusion gene under control of the CaMV35S promoter into the onion epidermal cells, we found that the OsLG1-GFP fusion protein localized to the nucleus (Fig. 5a), as predicted for a functional transcriptional factor. RNA in situ hybridization results showed that OsLG1 was strongly expressed at the panicle pulvinus (Fig. 5b), which is consistent with its role of controlling cellular development in that region.
Using quantitative RT-PCR, we found that OsLG1 mainly expressed at the panicle pulvinus, leaf sheath, ligule, leaf blade and culm (Fig. 5c). OsLG1 expression at the panicle pulvinus was higher in YIL31 and CIL49 than that in Teqing and C418 (Fig. 5c). To further investigate whether the OsLG1 gene expression changes result from the methylation level differences of the OsLG1 promoter region, we collected the panicle pulvinus of YIL31 and Teqing and performed bisulfite sequencing to compare DNA methylation levels of the 1.4-kb promoter region upstream the translation start site. Significant differences of methylation were observed at Cyt_241 and Cyt_244 cytosine site, with higher levels of methylation in Teqing (B83% at the Cyt_241 site, B40% at the Cyt_244 site) compared with YIL31 (B40% at the Cyt_241 site, B16% at the Cyt_244 site; Supplementary Fig. S3, Supplementary Table S3), implying that the DNA methylation levels in the promoter region of OsLG1 might affect the OsLG1 gene expression change between Teqing and YIL31 in the panicle pulvinus.

Discussion
In this paper, we demonstrated that the OsLG1 is a key gene controlling rice panicle architecture. Association analysis indicates that selection of an SNP residing in the cis-regulatory region of the OsLG1 gene was responsible for the transition from a spread panicle typical of ancestral wild rice to the compact panicle of present cultivars during rice domestication.
A rice leaf consists of a leaf blade, a leaf sheath and a laminar joint that contains a pair of auricles and the ligules. Previous study showed that OsLG1 regulated the rice ligule development by characterization of a T-DNA knockout mutant 18 . The observations of dual roles for OsLG1 in regulating ligule development and panicle architecture indicate that OsLG1 has pleiotropic effects. The ZmLG1 gene, presumably orthologous to OsLG1 (Fig. 4), also controls the branch angle of the tassel and is  required for the ligule development in maize 20,26 . Knockout mutants of ZmLG1 result in a more compact tassel than their wild-type counterparts and loss of the ligules in maize, indicating that the pleiotropic gene functions of both OsLG1 and ZmLG1 are conserved between these species. The nucleotide diversity of cultivated rice in the 3.3-kb regulatory region of the OsLG1 gene was significantly lower compared with the ancestral species, O. rufipogon (Fig. 3d). Very recently, a separate work also indicated that the upstream regulatory sequences of the OsLG1 gene were responsible for the compact panicle of domesticated rice, although the FM was not identified 27 . Furthermore, the whole-genome sequencing study showed that OsLG1 was under strong artificial selection 28 . These results suggested that the OsLG1 region experienced artificial selection during the rice domestication.
Although several genes including qSH1 and TB1, controlling domestication-related traits with the causal mutations in the cisregulatory region, had been identified 6,29-31 , the advantage of the selection of cis-regulatory mutation during crop domestication is poorly known. OsLG1 was shown to have pleiotropic effects with regard to spread panicle architecture and normal ligule development in rice 25,27 (Fig. 2). The ligule is an important functional organ as it can prevent rainwater and dust from entering the leaf sheath. Although a spread panicle is beneficial for wild rice, the compact panicle is a desirable trait for cultivation purposes. The domestication of rice, therefore, required a transition in panicle architecture, from spread to compact, without changing the ligule. Either intentionally or fortuitously, mutations in the OsLG1 regulatory region, reducing its expression specifically at the panicle pulvinus, were selected by ancient humans for a favourable compact panicle architecture in domesticated rice while maintaining normal ligule formation.
Crop domestication is the genetic modification of a wild species to generate a new form of a crop altered to meet human needs 32 . During domestication and improvement of crops, problematic agronomic traits associations often occur, which result from pleiotropic effects of the target gene on other traits, or tight linkage of genes that control independent traits (known as 'linkage drag'), which pose penalties on crop production 33 . Strategies targeting changes to the favourable gene function circumventing undesirable agronomic trait associations has been a difficult problem for crop breeders. Our findings indicate that sequence variance in the cis-regulatory region finely tune spatial expression levels of pleiotropic gene effects, possibly by changing the methylation status of the promoter region of the target gene, and consequently break the negative agronomic trait association resulting from pleiotropic gene effects. This novel discovery is very important for understanding the mechanism of crop domestication and will improve the molecular breeding of crops in the future.

Methods
Plant materials. Teqing, an elite indica cultivar (O. sativa L.), and YJCWR, a common wild-rice accession (O. rufipogon Griff.) collected as rhizomes from Yujiang province, China, were used as recurrent and donor parents in a backcrossing programme, respectively. The F 1 plant derived from the cross between Teqing and YJCWR were backcrossed three times consecutively with Teqing to generated introgression population. We selected an introgression line YIL31 displaying a spread panicle phenotype to backcross with Teqing. The F 2 -segregating population was used for genetic analysis and fine mapping. A set of 158 diverse O. sativa and 21 accessions of wild rice from 17 countries used in this study are listed in Supplementary Table S1.
Primers. The primers used in this study are listed in Supplementary Table S2.
Complementation test. The BAC clone YJ0510607, containing the OsLG1 gene, was identified from the BAC library of YJCWR 34 . The 11.6-kb and 13.7-kb fragments, harbouring the 3.3-kb-mapping region, were inserted into the binary vector pCAMBIA1300 to form pV1 and pV2, respectively. These two plasmid constructs were introduced to Agrobacterium tumefaciens strain LBA4404 and subsequently transferred into the japonica cultivar Zhonghua 17, which had compact panicle. The construct pV3 contained a 28.5-kb segment harbouring the whole fine-mapping region and the OsLG1 gene. The pV3 were mixed with pCAMBIA1300 and transferred into Zhonghua 17 using a helium biolistic device (Bio-Rad PDS-1000). To generate the overexpression construct pOE, the ORF of OsLG1 was amplified from first-strand cDNA of YIL31. The construct drove the expression of OsLG1 cDNA under the control of the ubiquitin promoter.
Subcellular localization. Subcellular localization of OsLG1 was determined using the coding sequence of a GFP fused in-frame to the OsLG1-coding sequence and transcribed from a CaMV35S promoter. The resulting plasmid was bombarded into onion epidermal cells using a helium biolistic device (Bio-Rad PDS-1000). We examined the bombarded tissues with a confocal laser-scanning microscope (Carl Zeiss LAM510).
RNA in situ hybridization experiment. The rice young panicles of YIL31 and Teqing were collected and fixed with 4% (w/v) paraformaldehyde at 4°C overnight, followed by a series of dehydration and infiltration, and were embedded in paraffin (Paraplast Plus, Sigma). The tissues were sliced into 8-mm sections with a microtome (Leica RM2145). The 396-bp 3 0 -region of OsLG1 cDNA was subcloned into the pSK vector and used as the template to generate sense and antisense RNA probes. Digoxigenin-labelled RNA probes were prepared using a DIG Northern Starter Kit (Catalogue number 2039672, Roche) according to the manufacturer's instruction. Slides were observed under bright field through a microscope (Leica DMR) and photographed with a micro colour charge-coupled device camera (Apogee Instruments).
Quantitative RT-PCR. We extracted total RNAs using an RNeasy Plant Mini Kit (Qiagen). First-stand cDNA synthesis was done with BcaBEST RNA PCR Kit (TaKaRa). Real-time RT-PCR was done on the ABI Prism 7900 Sequence Detection System (Applied Biosystems). Diluted cDNA was amplified using the SYBRGreen Master Mix (Applied Biosystems). We normalized the levels of OsLG1 transcripts by endogenous 18S rRNA transcripts amplified with primers 18SF and 18SR. Each set of experiments was repeated three times, and the relative quantification method (2 À DDC T (DDCT)) was used to evaluate quantitative variation.
Bisulfite genomic sequencing. We collected the base of panicle branch of YIL31 (containing the panicle pulvinus) and Teqing (without the panicle pulvinus), and then isolated the genomic DNA. The DNA was bisulfite treated using the EZ DNA Methylation-Gold kit. The 1.4-kb promoter region upstream the translation start site of OsLG1 gene was amplified using bisulfite primers and cloned into the pCR-4 vector. Each fragment was sequenced in at least 30 clones.
Association mapping. Association testing was performed with Fisher's exact test in R, as Fisher's exact test can be powerful for the binary trait such as rice panicle branch angle 35