Genome-wide association mapping reveals a rich

Full text

Turn on search term navigation

ARTICLE

Received 14 Jan 2011 | Accepted 2 Aug 2011 | Published 13 Sep 2011 DOI: 10.1038/ncomms1467

Genome-wide association mapping reveals a rich genetic architecture of complex traits in Oryza sativa

Keyan Zhao1,2, Chih-Wei Tung3, Georgia C. Eizenga4, Mark H. Wright1, M. Liakat Ali5, Adam H. Price6, Gareth J. Norton6, M. Raqul Islam7, Andy Reynolds1, Jason Mezey1, Anna M. McClung4, Carlos D. Bustamante1,2 & Susan R. McCouch3

Asian rice, Oryza sativa is a cultivated, inbreeding species that feeds over half of the world s population. Understanding the genetic basis of diverse physiological, developmental, and morphological traits provides the basis for improving yield, quality and sustainability of rice. Here we show the results of a genome-wide association study based on genotyping 44,100 SNP variants across 413 diverse accessions of O. sativa collected from 82 countries that were systematically phenotyped for 34 traits. Using cross-population-based mapping strategies, we identied dozens of common variants inuencing numerous complex traits. Signicant heterogeneity was observed in the genetic architecture associated with subpopulation structure and response to environment. This work establishes an open-source translational research platform for genome-wide association studies in rice that directly links molecular variation in genes and metabolic pathways with the germplasm resources needed to accelerate varietal development and crop improvement.

1 Department of Biological Statistics and Computational Biology, Cornell University , Ithaca , New York 14850 , USA . 2 Department of Genetics, Stanford University, Stanford, California 94305, USA.3 Department of Plant Breeding and Genetics, Cornell University, Ithaca, New York 14850, USA.4 USDA ARS, Dale Bumpers National Rice Research Center, Stuttgart, Arkansas 72160, USA.5 Rice Research and Extension Center, University of Arkansas , Stuttgart , Arkansas 72160, USA.6 Institute of Biological and Environmental Sciences, University of Aberdeen, Aberdeen AB24 3UU, UK.7 Department of Soil Science, Bangladesh Agricultural University , Mymensingh 2202 , Bangladesh . Correspondence and requests for materials should be addressed to S.R.M.(email: [email protected] ) or to C.D.B. (email: [email protected] ) .

NATURE COMMUNICATIONS | 2:467 | DOI: 10.1038/ncomms1467 | www.nature.com/naturecommunications

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms1467

Understanding the genetic basis of physiological, developmental and morphological variation in domesticated Asian rice ( Oryza sativa) is critical for improving the quality,

safety, reliability and sustainability of the world s food supply. Human population growth, particularly in developing countries where rice is the main source of caloric intake 1, coupled with climate change and the intensive water, land and labour requirements of rice cultivation 2 , creates a pressing and continuous global need for new, stress tolerant, resource-use effi cient, and highly productive rice varieties. To assist in this endeavour, the scientic community has created a wealth of genomic and plant breeding resources, including high-quality genome sequences 3,4 , dense SNP maps 57, extensive germplasm collections 6,8,9 and public databases of genomic information 5,6,10,11.

Despite the availability of these scientic resources, most of what we know about the genetic architecture of complex traits in rice is based on traditional quantitative trait locus (QTL) linkage mapping using bi-parental populations. While providing valuable insights 12 , the QTL approach is clearly not scalable to investigate the genomic potential and tremendous phenotypic variation of the more than 120,000 accessions available in public germplasm repositories. Genome-wide association study (GWAS) mapping makes it possible to simultaneously screen a very large number of accessions for genetic variation underlying diverse complex traits. An extra advantage of the GWAS design for rice is the homozygous

nature of most rice varieties, which makes it possible to employ a genotype or sequence once and phenotype many times over strategy, whereby once the lines are genomically characterized, the genetic data can be reused many times over across dierent phenotypes and environments.

Here we present a genome-wide association study in a global collection of 413 diverse rice ( O. sativa ) varieties from 82 countries using a high-quality custom-designed 44,100 oligonucleotide genotyping array. For these varieties, we systematically phenotyped 34 morphological, developmental and agronomic traits over two consecutive eld seasons. Our mapping strategy evaluated variation both within and among four of the major subgroups of rice, revealing signicant heterogeneity of genetic architecture among groups, as well as gene-by-environment eects. Unlike previous GWAS studies in rice 5,puried seed stocks of the rice strains and all the genotypic and phenotypic information generated over the course of this study are publicly available, creating a valuable, open-source translational research platform that can be rapidly expanded through community participation to enhance the power and resolution of GWAS in rice.

ResultsDiversity panel and genotyping array. A rice diversity panel consisting of 413 inbred accessions of O. sativa collected from 82 countries ( Fig. 1 ; Supplementary Data 1 ) was genotyped using an

Aus (57)

Indica (87)

Aromatic(14)

Temperate japonica (96)

Admixed(62)

Tropical japonica(97)

Indica

Aus Temperate japonica

Aromatic Tropical japonica

Aus Temperate japonica

Aromatic Tropical japonica

0.25

0.20

0.15

0.10

0.05

0.00

0.05

0.04

PC1 (34.3%)

0.06

0.02

0.00

0.02

0.04

PC4 (2.3%)

0.05 0.00 0.05 0.10 0.10 0.05 0.00 0.05

PC2 (9.8%) PC3 (5.9%)

Figure 1 | Population structure in

O. sativa. (a) The large pie chart summarizes the distribution of subpopulations in the 413 O. sativa samples in our diversity panel, and the smaller pie charts on the world map correspond to the country-specic distribution of subpopulations sampled (note: large countries such as China, India and the US were divided into several major rice growing regions). The size of the pie chart is proportional to the sample size and colours within each pie chart are reective of the percentage of samples in each subpopulation. Seeds representing each subpopulation are displayed with and without hull in the centre, with 1 cm scale bar. ( b) Principal component analysis was used to provide a statistical summary of the genetic data, and the top four principle components are illustrated in the bottom panels.

NATURE COMMUNICATIONS | 2:467 | DOI: 10.1038/ncomms1467 | www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms1467

ARTICLE

Table 1 | Polymorphism summary of Affymetrix 44K SNPs in each subpopulation.

Aus Indica Tropical japonica Temperate japonica Aromatic/GroupV

Private SNPs 822 1,851 398 376 77 Polymorphic SNPs 23,270 30,449 24,813 14,688 12,059 MAF>=0.05 18,012 20,259 13,051 7,775 12,039

Private SNPs are unique to one specic subpopulation; Polymorphic SNPs are considered to be those that segregated in one specic subpopulation, irrespective of whether they also segregate in another subpopulation (they could also be polymorphic or xed in other subpopulations). MAF, minor allele frequency.

Aymetrix single nucleotide polymorphism (SNP) array containing 44,100 SNPs (herea er referred to as the 44 K chip). With a genome size of ~380Mb (ref. 13), this custom-designed genotyping chip provides high quality data (less than 4.5 % missing data), with ~ 1 SNP per 10 kb across the 12 chromosomes of rice (Methods; Supplementary Data 2 ). Th e diversity panel was evaluated for 34 traits related to plant morphology, grain quality, plant development and agronomic performance using eld-grown plants with replications within and between years ( Supplementary Table S1 ; Supplementary Data 3).

Population structure and linkage disequilibrium estimation in rice. Using principle component analysis (PCA)14 to summarize global genetic variation in the diversity panel, we observed clear, deep subpopulation structure in this collection of germplasm (Fig. 1a). The top four principal components (PCs) explained almost half of the genetic variation ( Fig. 1b ). The ve subpopulations indica, aus, temperate japonica, tropical japonica and aromatic formed clear clusters based on the top four PCs, and were well dierentiated from each other, with pairwise Fst (F-statistic) values ranging from 0.23 0.53. Th is is in agreement with previous ndings where global germplasm collections have been used in combination with much smaller numbers of SNP or simple sequence repeat (SSR) genotypes 8,1517. Because the array was designed to assay variation in all O. sativa groups, most SNPs are shared or polymorphic across subpopulations ( Table 1 ).

We examined allele sharing across the panel by calculating identity by state coeffi cients among all pairs of accessions ( Fig. 2a ). We nd that whereas allele sharing clearly tracks subpopulation ancestry as identied by the PCA analysis, there is also a substantial number of admixed accessions, highlighting the complex history of rice varieties grown throughout the world 16. Excluding the small sample of aromatic accessions, the mean observed identical by state (IBS) sharing is greatest between the closely related tropical japonica and temperate japonica accessions (0.80), followed by indica and aus (0.64), with relatively little IBS sharing between the two major subspecies, Indica and Japonica (0.47) ( Fig. 2a ). Th e fact that most of the admixture occurs within (rather than between) subspecies underscores the existence of genetic and cultural barriers to genetic exchange between these two major groups of Asian rice, despite documented cases of targeted Japonica-Indica introgression mediated by articial selection 18,19.

Th e amount of genomic variation tagged by our SNP array was calculated by measuring the pairwise SNP linkage disequilibrium (LD) among the 44 K common SNPs. On average, LD drops to almost background levels around 500 kb 1 Mb, reaching half of its initial value at ~ 100 kb in indica, 200 kb in aus and temperate japonica, and 300 kb in tropical japonica ( Supplementary Fig. S1 ). Given that our average inter-marker distance is 10 kb, we expect to have reasonable power to identify common variants of large eect associated with our traits of interest, even if we have not queried the causal variant for association in the domesticated varieties.

Phenotypic variation. Th e phenotypes we examined in our GWAS can be classied broadly into six categories: plant morphology-

related traits; yield-related traits; seed and grain morphology-related traits; stress-related phenotypes; cooking, eating and nutritional-quality-related traits; and plant development, represented by owering time, which we measured in three geographic locations that diered in day-length and ambient temperature. Canonical correlation analysis demonstrated that phenotypes within a category are oen correlated, ranging from a low of 0.41 between brown rice seed width and brown rice seed length, to a high of 0.9 between hulled and dehulled seed morphology ( Fig. 2b ;

Supplementary Fig. S2 ).

For all the phenotypes evaluated in this study, we observed global similarities among members of the same subpopulation, consistent with the domestication and breeding history of these varieties. Correlation coeffi cients between accession pairs across all phenotypes were signicantly higher for accession pairs from the same subpopulation than from dierent subpopulations ( P<2.2e16, one-sided MannWhitney U -test) (lower triangle of Fig. 2a ). Consistent with this observation, the top four PCs (based on the 44 K SNPs mentioned above) explained a large proportion of phenotypic variation, with values ranging from 20 40 % ( Supple mentary Table S1 ). In the case of rice grain, morphological and cooking-quality traits are key to varietal identity and have been under strong diversifying selection by humans in dierent parts of the world 1821 . Physical grain characteristics in rice are salient because they serve as indicators of local and regional eating prefe rences in a crop that, unlike wheat or maize, is consumed largely as whole kernel. Traits such as owering time and disease resistance are also strongly correlated with region and environment, meaning that genotypic, phenotypic and environmental variation in O. sativa are all correlated to some degree, posing signicant challenges for GWAS.

The strong confounding eect of population structure. The results of our genome-wide association scans are summarized in Supplementary Figures S3 S36 where we show SNP-trait associations discovered in the diversity panel as a whole, as well as in each subpopulation individually. As can be seen in the quantile quantile plots (Fig. 3a; Supplementary Figs. S3S36), the distribution of observed log10 P -values from the na ve analysis (no population structure adjustment) departed quite far from the expected distribution under a model of no association (that is, the P-values should lie on the diagonal line), with signicant ination of nominal P -values leading to a high level of false positive signals. Use of a modied mixed model strategy 2224 allowed us to consider dierent levels of population structure and relatedness in our diversity panel. This eectively eliminated the excess of low P-values for most traits, but it also likely eliminated true positives. Th is is a common problem seen in other systems as well; for example, geographic coordinates correlate closely with owering time in plants 24.

For this reason, we believe a combination of na ve and population structure-adjusted hits, coupled with subpopulation-specic analyses in rice, is the most thoughtful way to identify potential variants for follow up.

Using the mixed model 23 to analyse the associations between 34 phenotypes and 44 K SNP genotypes evaluated in our 413 O. sativa rice lines, we successfully identied both known associations (for example, enrichment in a priori candidate genes and

NATURE COMMUNICATIONS | 2:467 | DOI: 10.1038/ncomms1467 | www.nature.com/naturecommunications

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms1467

0.8

0.6

0.4

0.2

0.4

0.6

0.8

Tropicaljaponica Temperate

japonica Aromatic

Indica

Aus

Flowering time at Arkansas

Flowering time at Faridpur

Flowering time at Aberdeen

FT ratio of Arkansas/Aberdeen

FT ratio of Faridpur/Aberdeen

Culm habit

Leaf pubescence

Flag leaf length

Flag leaf width

Awn presence

Panicle number per plant

Plant height

Panicle length

Primary panicle branch number

Seed number per panicle

Florets per panicle

Panicle fertility

Seed length

Seed width

Seed volume

Seed surface area

Brown rice seed length

Brown rice seed width

Brown rice surface area

Brown rice volume

Seed length/width ratio

Brown rice length/width ratio

Seed color

Pericarp color

Straighthead susceptibility

Blast resistance

Amylose content

Alkali spreading value

Protein content

Flowering time

Morphology

Yield components

Seed morphology

Stress tolerance

Quality

Max

Min

Figure 2 | Identity by State and phenotypic variation among subpopulations. ( a) Individuals are ordered according to their genotypic distance(1-IBS, identied by state) clustering with the tree shown on the right. The upper diagonal shows the IBS-sharing between individuals (values rescaled from 0 to 1). The lower diagonal shows the individual correlation coefcients based on all phenotypes. Coloured bars along the bottom of the panel reect the sample subpopulation assignment as labelled; dark colour within each subpopulation indicates admixed individuals. ( b) Summary of phenotypic distributions among all individuals, with phenotypes grouped by trait category and individuals grouped by subpopulation as in ( a).

previously reported QTLs from rice and other species) as well as new candidate loci in the rice genome. Detailed results for each of the 34 phenotypes can be found in the Supplementary Data 3 as well as online in the Gramene database ( www.gramene.org ) and on our project website (www.ricediversity.org/44kgwas).

Trade os between the mixed model and na ve model. Plant height is an important developmental and yield-related trait. Dozens of genes regulating plant height in rice have been identied previously including dwarng mutants 25, QTLs12, orthologues from other plant species, and genomic targets of ne-mapping experiments related to harvest index and yield 12,26 . Both the na ve and the mixed model

consistently detected strong signal linked to the Green Revolution semi-dwarf gene, SD1 , on chromosome 1 ( Fig. 3d ). Interestingly, several SNPs near other height-controlling genes such as OsBAK1 on chromosome 8 (ref. 27), DGL1 on chromosome 1 (ref. 28 ) were only detected by the na ve approach ( Fig. 3d ). Th is suggests that, in the case of rice, the mixed model may overcompensate for population structure and relatedness, leading to false negatives. Therefore, the many mapping resources derived from crosses between parents belonging to dierent subpopulations and Oryza species will be needed to complement GWAS, helping to reduce the rate of false positives and false negatives 24 , yielding QTLs that cannot be identied by mapping within subpopulations 29.

NATURE COMMUNICATIONS | 2:467 | DOI: 10.1038/ncomms1467 | www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms1467

ARTICLE

b c

180

160

140

Observed log 10(P)

Frequency

120

100

0 1 2 3 4 5

Expected log10(P)

Admix

Aromatic

Aus

Ind

Tej

Trj

60 80 100 120

140 160 180 200

SD1

DGL1

LAX1

Mixed model log 10(P)

Naive log 10(P)

DLT

OsBAK1

SD1

2 3 4 5 6 7 8 9 10 11 12

1 2 3 4 5 6 7 8 9 10 11 12

Chromosome

Figure 3 | Phenotypic distribution and genome-wide association scan for plant height. ( a) Quantile Quantile plots for both na ve and mixed model for plant height in all samples. ( b) Boxplot showing the differences in plant height among subpopulations. Box edges represent the upper and lower quantile with median value shown as bold line in the middle of the box. Whiskers represent 1.5 times the quantile of the data. Individuals falling outside the range of the whiskers shown as open dots. ( c) Histogram of plant height in all samples. Dashed black line represents the null distribution. ( d) Genome-wide P-values from the mixed model and na ve method. x axis shows the SNPs along each chromosome; y axis is thelog10 (P-value) for the association.

Coloured dots in ( a) and ( c) indicate SNPs with P-values<1104 in the mixed model and the top 50 SNPs in the na ve method; SNPs within 200 kb range of known genes are in red; other signicant SNPs are in blue. Candidate gene locations shown as red vertical dashed lines with names on top.

Genetic heterogeneity across subpopulations. In our diversity panel, aromatic varieties had the longest mean panicles (30 cm), temperate japonica had the shortest (21

cm), aus and indica had intermediate panicle length, and the greatest range of panicle length was observed among tropical japonica varieties ( Fig. 4a ).

To determine whether dierent networks of alleles were associated with trait variation in the dierent subpopulations, we per formed GWAS on each subpopulation independently and in the panel as a whole, and compared results. As summarized in Figure 4a,b , the genetic architecture of panicle length diers signi cantly among subpopulations and dierent GWAS peaks are observed when the subpopulations are analysed individually or when the diversity panel is analysed as a whole. For example, in the indica population, we see clusters of highly signicant SNPs near OsTB1[TEOSINTE BRANCHED1(ref.30)],

SLR1[SLENDER RICE1

OsBRI1 [syn. DWARF61, or

BRASSINOSTEROID-INSENSITIVE1 (ref. 32 )], in the aus subpopulation we observe signicant SNPs near FZP [FRIZZY PANICLE33] and SSD1 [SWORD SHAPE DWARF1 (ref. 34 )], and in the tropical japonica population, we see SNPs near OsLIC [LEAF AND TILLER ANGLE INCREASED CONTROLLER35] and MOC1 (MONOCLUM 1 (ref. 36)).

From these results, we conclude that dierent networks of genes regulate panicle length in dierent subpopulations and propose that subpopulation-derived genetic heterogeneity is a general pattern in O. sativa. Th is suggests that the Indica and

Japonica varietal groups should be properly treated as true sub-species for association analyses, and helps explain why crosses between members of divergent subpopulations, as well as between cultivated and wild species, oen give rise to transgressive o-spring 37 . We also demonstrate that the subpopulations of O. sativa contain alleles with vastly dierent eect-size on many traits of interest (that is, allele eects that are in the opposite direction to mean subpopulation dierences for those traits). This conforms to the general mechanism that explains the production of extreme, or transgressive, phenotypes at both the species level and below 37,38 and suggests a blueprint for harnessing natural variation to liberate transgressive phenotypes in the context of plant improvement.

Genotype by environment eects. To investigate how environmental variation aected the performance of GWAS, we evaluated owering time in three dierent environments and compared results. One experiment was conducted during 2007 in the eld in Stuttgart, Arkansas, USA (34 4 ) under long-day conditions (~1412h during May September); one was conducted in the eld in Faridpur, Bangladesh (235) under ~1213h days (JanuaryMay); and the third was conducted in the greenhouse in Aberdeen, Scotland, UK (579 ) across a nine-month period during which the days became very long and then very short (a range of ~ 18 6 h during the period spanning March December). Th e GWAS peaks explained

(ref. 31)] and

NATURE COMMUNICATIONS | 2:467 | DOI: 10.1038/ncomms1467 | www.nature.com/naturecommunications

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms1467

Admix

AT5G59570

TOC1

OsMADS51

OsPRR1

AT3G10185

Hd1

OsMADS13

Frequency

CKL13

LSH1

CRY2

DDF2

HAT4

COL2

CLF

COL9

GIS

AGL12

SHP1

Aberdeen

20 25 30 cm

Aromatic

Aus

Ind

log 10(P)

OsBRI1

Tej

Trj

Arkansas

Faridpur

SSD1

OsTB1

SLR1

D35

MOC1

OsLIC

FZP

1 2 3 4 5 6 7 8 9 10 11 12

Chromosome

log 10(P)

0 6 4 2 0 6 4 2 06 4 2 0

6 All

Trj

Tej

Ind

Aus

4 2 0 6 4 2

AT3G10185

CoL2

CLF

TIC

PMI15

Hd1

AT5G28450

GIS

PGI1

CR88

Faridpur/Aberdeen

Arkansas/Aberdeen

5 4

1 2 3 4 5 6 7 8 9 10 11 12

Chromosome

Figure 4 | Genetic heterogeneity of panicle length across subpopulations. ( a) Histogram showing distribution of panicle length across the diversity panel and boxplot showing differences in panicle length among subpopulations. In boxplot, the box edges represent the upperand lower quantile with median value shown as bold line in the middle of the box. Whiskers represent 1.5 times the quantile of the data. Individuals outside of the range of the whiskers shown as open dots.( b) Genome-wide P-values from the mixed model for panicle length forall 413 accessions in top panel ( all), and for tropical japonica, temperate japonica, indica and aus subpopulations individually in subsequent panels. Note: the aromatic subpopulation was not included because of the small sample size. X-axis indicates the SNP location along the 12 chromosomes;y axis is thelog10 (P value) from each method. Coloured dots indicate

SNPs with P-values<1104 in the mixed model; SNPs within 200 kb range of known genes are in red; other signicant SNPs are in blue. Candidate genes near peak SNP regions known to be previously associated with panicle, stem and internode elongation in rice are shown along the top.

Chromosome

Figure 5 | Genome-wide association scan for owering time.( a) Genome-wide P-values from the mixed model for owering time in three geographic locations are shown in the three panels. Association analysis in each subpopulation is shown in each row of the matrix. X axis indicates the SNP location along the 12 chromosomes, with chromosomes separated by vertical grey lines; y axis is thelog10 (P value) from each method. Candidate genes previously shown to determine owering time near peak SNPs are shown along the top, rice genes are in red, Arabidopsis homologues are in black. SNPs with P value<1104 are indicated by coloured dots. SNPs within 200 kb range of known rice owering time genes are in red; SNPs within 200 kb range of Arabidopsis owering-time homologues are in magenta; other signicant SNPs are in blue. ( b) GWAS regions associated with photoperiod sensitivity, calculated as the ratio of days-to-owering across pairs of environments.

between 5 50 % of the phenotypic variation for owering time in each environment ( Supplementary Data 4 ). As seen in Figure 5a , 10 genomic regions were associated with candidate genes for owering time under one or more daylengths while only the HEADING DATE 1 (HD1 ) region on chromosome 6 was detected in more than one environment.

Th e most signicant signal was observed under very long days in Aberdeen around HD1 , the major photoperiod-sensitivity locus, (synonym: SE1, or OsCONSTANS,

OsCO) on chromosome 6 (ref. 39 ). A well-dened peak in the same location was observed under long days in Stuttgart, AR when either the entire diversity panel or the temperate japonica subpopulation was analysed. The signicant SNPs detected in Aberdeen covered an extensive region of ~2.3Mb around

HD1, corresponding to a mountain range as described by Atwell et al.40 Th e mountain range distribution may

be due to the presence of several linked genes that contribute to owering time across the region, and / or to the presence of multiple alleles at the HD1 locus, along with multiple introgression events that have been documented within a 5.5

Mb region around the HD1 gene 41 . In domesticated species like rice, loci that are critical to both local adaptation and yield performance are o en the targets of both natural and articial selection, leading to complex forms of allele sharing and admixture in diverse varieties.

Some varieties were highly sensitive to daylength and others, mostly temperate japonica accessions, were insensitive to photo-period and owered at similar times across the three environ-

NATURE COMMUNICATIONS | 2:467 | DOI: 10.1038/ncomms1467 | www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms1467

ARTICLE

123.323.4

140.440.4

14343

217.417.7

219.919.9

224.624.7

33.53.8

428.728.7

61.91.9

64.64.6

67.88.3

68.48.8

68.99.4

69.49.9

61010.1

610.811.1

61313

614.914.9

616.416.4

619.819.8

723.223.3

822.922.9

823.724.1

824.224.2

921.821.9

104.84.8

1020.420.4

124.95.1

125.55.6

10.50.5

13.43.5

17.27.2

129.429.4

26.76.7

223.123.1

224.324.7

226.326.3

234.434.4

35.55.5

327.627.6

416.416.5

417.417.4

422.522.5

430.930.9

432.132.1

50.81.2

533

56.86.8

610.410.4

627.227.2

719.519.5

720.120.1

724.824.8

726.626.6

824.524.5

826.226.7

99.29.2

910.910.9

918.718.7

1019.519.5

112.62.6

119.19.1

1111.411.4

1124.324.6

122.42.4

1213.714

1221.521.5

1227.127.2

Flowering time at Arkansas

Flowering time at Faridpur

Flowering time at Aberdeen

FT ratio of Arkansas/Aberdeen

FT ratio of Faridpur/Aberdeen

Culm habit

Leaf pubescence

Flag leaf length

Flag leaf width

Awn presence

12.32.3

113.813.8

131.731.7

133.133.2

138.138.5

26.86.8

29.59.5

226.326.3

30.90.9

322.723

323.223.5

432.732.7

620.720.8

622.222.2

83.83.8

919.719.8

921.421.9

1128.328.3

123.73.7

1224.724.7

13.13.1

133.233.2

21818

221.721.7

224.524.5

229.629.9

314.214.2

315.816.3

316.416.8

317.117.2

317.718

318.818.8

321.221.7

322.122.5

322.623.1

323.223.5

327.627.7

42.52.5

42929

511

54.85.3

55.35.5

523.624

528.228.4

63.63.6

617.217.2

622.322.3

62323

75.96.1

722.122.1

724.824.8

815.715.7

824.224.2

95.95.9

106.16.1

1020.420.4

1021.721.7

1022.622.6

111.11.2

113.63.6

111717

1124.324.3

1125.825.8

1217.717.7

1219.119.1

1221.621.7

1225.225.5

Panicle number per plant

Plant height

Panicle length

Primary panicle branch number

Seed number per panicle

Florets per panicle

Panicle fertility

Seed length

Seed width

Seed volume

Seed surface area

Brown rice seed length

Brown rice seed width

Brown rice surface area

Brown rice volume

Seed length/width ratio

Brown rice length/width ratio

Seed colour

Pericarp colour

127.227.2

229.729.7

235.135.1

37.27.2

328.828.8

53.53.5

54.95

60.90.9

61.61.8

63.23.6

64.24.4

655

65.95.9

66.56.9

677

67.67.9

71.61.6

719.520

726.626.6

72929

81.61.6

84.75

922.422.6

106.16.1

1015.415.4

1217.117.4

1218.818.9

431.431.4

64.64.6

114.44.4

1121.821.8

1210.210.6

1210.810.9

Amylose content

Alkali spreading value

Protein content

Straighthead susceptibility

Blast resistance

No significant associations1e 5 <= P < 1e4 1e 6 <= P < 1e5 1e 7 <= P < 1e6 P < 1e7

qSW5

Maximum effect locus All significant loci

Contribution to phenotypic

variance (%)

CYP78A5

GS3

Waxy

SSII-3

qSW5

SD1

Pi-ta

OsMADS13 Hd1

D35

GS3

Hd1

12 Hd1

Flowering time at Arkansas

Flowering time at Faridpur

Flowering time at Aberdeen

FT ratio of Arkansas/Aberdeen

FT ratio of Faridpur/Aberdeen

Flag leaf length

Flag leaf width

Panicle number per plant

Plant height

Panicle length

Primary panicle branch number

Seed number per panicle

Florets per panicle

Panicle fertility

Brown rice volume

Seed length/width ratio

Brown rice length/width ratio

Blast resistance

Amylose content

Alkali spreading value

Protein content

Seed length

Seed width

Seed volume

Seed surface area

Brown rice seed length

Brown rice seed width

Brown rice surface area

Figure 6 | Summary of trait associations across genomic regions and percentage of variance explained by signicant locus. ( a) Each row representsa trait, and each column corresponds to a genomic region containing multiple SNPs that are signicantly associated with a trait. Signicance is colour-coded based on the P value of the association. ( b) The x axis represents the trait, the y axis shows the contribution ( % ) of signicant loci. Candidate genes detected within 200 Kb region of signicant loci are labelled on top of the maximum effect locus.

ments. When photosensitivity (the ratio of days-to-owering across pairs of environments) was used as a derived trait for GWAS, the most signicant SNPs ( P<10 6 ) were consistently found near HD1 , followed by a region on chromosome 7 containing homo-logues of Arabidopsis genes known to regulate the circadian rhythm [that is, TIME FOR COFFEE, TIC42 ] and light sensing [ PLASTIC MOVEMENT IMPAIRED 15, PMI1543](Fig.5b),whichhave not previously been shown to be associated with natural variation for owering time in rice.

We also demonstrate the eect of genotype by environment (GxE) interaction for owering time by comparing GWAS results over two years in the same location in Stuttgart, Arkansas ( Supplementary Fig. S37 ). In this case, we observe extensive year-to-year variation between 2006 and 2007, mostly due to the dierent weather patterns experienced during the two growing seasons. Several GWAS peaks associated with candidate genes are signicant in only a single year. The HD1

locus was signicant in 2007, but not in 2006, though it was signicant using the average owering time from two years.

Although the genetic complexity and low heritability of owering time as well as several other traits evaluated here in eld grown plants ( Supplementary Figs S37 S40 ) tend to confound the interpretation of GWAS results, this study provides an opportunity to

look carefully at a range of ecologically and agronomically important traits evaluated under natural growing conditions and compare GWAS results with prior QTL and mutant studies to better understand plant growth and development 44,45.

Gene linkage or pleiotropy. A matrix summarizing the QTL regions associated with all traits, as well as the percent of the phenotypic variation explained by signicant SNPs for each trait, can be found in Figure 6 . For many traits, the maximum-eect locus falls within a 200 kb region containing a previously identied functional gene (as highlighted in Fig. 6b and summarized in Supplementary Data 4 ). When our results are compared with those of Huang et al.5,

the same known genes showed clear signal for the same phenotypes (for example, GS3 and qSW5 for grain length and width, SSII-3 and Waxy for alkaline spreading value and amylose content, Rc for pericarp colour). The signicant SNPs in our study explained up to 58 % of the variance compared with values up to 68 % reported by Huang et al.5 In addition, we evaluated traits not previously documented and identied known genes associated with those traits (for example, SD1 for plant height, OsMADS13 forowering time and Pi-ta for blast resistance). Th is demonstrates that our 44 K SNP array is capable of capturing the major common variants responsible for critical agronomic traits.

NATURE COMMUNICATIONS | 2:467 | DOI: 10.1038/ncomms1467 | www.nature.com/naturecommunications

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms1467

In several cases, the same SNPs were signicantly associated with multiple traits. Th is could be the result of pleiotropy or closely linked genes (local LD) 46. For example, we observed SNPs at 31 Mb chromosome 4 that were signicantly associated with both rice blast disease resistance and ag leaf width, and SNPs associated with rice blast disease resistance, amylose content and owering time at 4.2 4.6 Mb on chromosome 6. Th ese associations were also supported by Canonical correlation analysis based on traits measured in Arkansas ( Supplementary Fig. S2 , r= 0.3 for blast resistance and ag leaf width, r= 0.31 for blast resistance and owering time, r=0.37 for amylose content and owering time). Similar trait associations have been previously reported in these and other regions in rice 4749 . Linkage among favourable alleles is a strong determinant of phenotypic value under both natural and articial selection, a fact long appreciated by plant breeders. Validation studies involving joint linkage and association mapping, coupled with ne-mapping to identify the exact genes and alleles underlying our GWAS hits, will be required to more clearly understand the relationship between these candidate genes and the phenotypes observed in our panel 46 , as well as to provide breeders with the appropriate genomic tools needed to break deleterious linkages and liberate valuable alleles in this region.

Discussion

Th e deep population structure of O. sativa and its importance in explaining the heterogeneity of genetic architecture associated with most complex traits in rice underscores the value of using a worldwide diversity panel to untangle the genotype phenotype associations in the species. As demonstrated by our study, no single GWAS design or analysis method is suffi cient to unravel the complex genetics underlying natural variation in O. sativa. Th e nave approach has high false positive rates, and, although the mixed model successfully reduces ination of P-values, it o en masks true QTLs that are strongly correlated with population structure. In cases where alleles segregate across multiple subpopulations, the mixed model has the best power to nd them. However, when alleles segregate in only one subpopulation, or totally dierent alleles are present in dierent subpopulations, the na ve approach detects strong signals in the cloud of other, false signals, while the mixed model approach misses them entirely. As demonstrated by the IBS and Fst estimates, both divergence and heterogeneity among subpopulations is characteristic of the genomic pattern observed in rice. Subdividing the diversity panel to analyse subpopulations independently, using the mixed model, appears to provide a reasonable solution to this problem.

Given our marker density and sample size, this study is adequately powered to nd alleles of large eect that are common across populations, but a larger panel coupled with higher density of SNPs would empower us to detect more QTLs of small eects. It is noteworthy that some of the strongest signals are quite far from known candidate genes. Th is may be due, in part, to ascertainment bias where our best tag-SNP for a candidate gene is relatively far from the predicted locus, or we may be tagging previously undiscovered loci that happen to map near a known candidate. SNPs in high LD and with similar allele frequencies would give similar P-valuesin association. Th e SNPs used in our study were discovered by array-based re-sequencing of 20 O. sativa accessionsacross~100Mbof the genome 6 . Genetic variation discovered from deep next-generation sequencing in a larger number of accessions is likely to provide improved estimates of LD decay and more highly resolved views of local LD patterns in each subpopulation. Likewise, the integration of transcriptome data will improve our ability to detect moderate strength and rare alleles, as well as to begin to dissect the GxE eects and provide better resolution for the hits found in this study. Recent work by Nicolae 50 suggests that many trait-associated SNPs are likely to be eQTLs, and, in the case of owering time, there is abundant molecular evidence showing that gene expression

levels contribute directly to trait variation 44. Th us, the trajectory of GWAS in rice is similar to advances in human genetics, where initial studies employed several hundred and then thousands of individuals for common alleles, and subsequent work has been necessary to nd associations with either rare alleles or alleles of smaller eect 51.

Our results demonstrate that dierent traits have dierent genetic architectures. This reects the relative strength of environmental and human selection, with corresponding impacts on the phenotypic contribution of maximum eect and the total number of signicant SNPs. In some cases, a few genes in a pathway may lead to major changes in adaption, such as HD1 . In other cases, humans may exert selection in dierent directions on the same gene(s), such as seed length ( GS3)52 amylose content 21, and aroma 19. Where domestication-related loci are involved, we oen see SNPs with large eect that are shared across dierent populations 53,and

while they clearly distinguish O. sativa from its wild ancestors, these SNPs of large eect are oen diffi cult to detect in O. sativa, because they are nearly xed in cultivated material. Other SNPs, even those with only small eects, may be clearly identiable within individual populations. The subpopulation-specic allele distribution explains why crossing wild and domesticated rice, or one subpopulation with another results in transgressive variation in the progeny 37.

Both linkage drag and pleiotropic eects of a target gene can be either benecial or troublesome in the context of plant breeding 54,55 , and it is helpful to understand the underlying genetic cause of multiple trait associations. In the case of blast resistance, many late-maturing, tropical indica varieties that are resistant to blast disease are used as donors to introduce disease resistance into susceptible, early maturing temperate japonica varieties16.

However, undesirable traits such as late owering or inappropriate grain quality, may be co-introduced along with the disease resistance 48.Th e use of a broad diversity panel in GWAS not only serves to map associations between traits and DNA polymorphisms but also allows us to unravel the origin of genetic correlations among phenotypic traits, that is, pleiotropy versus genetically linked genes, and facilitates the selection of donors with combinations of traits that are likely to be adaptive and selectively advantageous for breeding in target environments.

We note that the rice diversity panel presented here represents an immortalized germplasm resource that is accompanied by both genotypic and phenotypic information ( Supplementary Figs S3 S36).Th e seeds are publicly available through the Genetic Stocks Oryza center in Stuttgart, AR ( http://www.ars.usda.gov/Main/docs. htm?docid=8318 ) or the International Rice Germplasm Collection at International Rice Research Institute in the Philippines ( http:// irri.org/our-science/genetic-diversity/get-and/or-submit-seeds). Th is enables people around the world to leverage the results of this project as the basis for continued association mapping without incurring any genotyping expense. The puried lines from this study can be used to generate MAGIC or NAM populations 56 to validate GWAS results and to further dissect the complex interaction among genes and environments that underlies quantitative variation in rice. Th e genotypic dataset and information about the 44 K SNP chip are publicly available ( www.ricediversity.org/44kgwas and www. gramene.org ) and can be used to design more targeted SNP assays for immediate use in variety identication, seed-purity testing, linkage analysis, pedigree conrmation and molecular breeding 57,58.

Our work highlights experimental design strategies and challenges involved in nding genes underlying phenotypic variation and is relevant to other species initiating GWAS, especially those with deep population structure. By launching this GWAS platform, we aim to deepen our understanding of natural variation and its phenotypic consequences, and to open the door to more effi cient utilization of the enormous wealth of diversity available in rice germplasm repositories around the world.

NATURE COMMUNICATIONS | 2:467 | DOI: 10.1038/ncomms1467 | www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms1467

ARTICLE

Methods

SNP array development and SNP selection. We selected 44,100 SNPs from 2 data sources: SNPs from the OryzaSNP project, an oligomer array-based re-sequencing eort using Perlegen Sciences technology 6 and BAC clone Sanger sequencingof wild species from OMAP project 59. Priority was given to SNPs with the least amount of missing data across the 20 SNP discovery accessions in the OryzaSNP project. SNPs were selected to tag all 159,879 high quality SNPs in the OryzaSNP data (Intersection set) with criteria of r2=1 and a conservative tagging window size of 50 kb. To further lter the SNPs, Blast was performed using the 33 bp sequence anking each SNP to remove any SNPs that mapped to more than 1 location inthe genome with fewer than 2 mismatches. Also, SNP targets were removed if there were SNPs detected within 15 bp of the target in the low-quality Union set (359,000 SNPs) in the OryzaSNP dataset. Th is yielded 31,663 tagging SNPs. We then selected 8,437 SNPs from a pool of SNPs from the Intersection and Union sets of the OryzaSNP data and another 4,000 SNPs from the OMAP dataset to ll in any gaps>20Kb between the tagging SNPs. Th is generated a well-distributed SNP array providing ~ 1 SNP every 10 kb along the 12 chromosomes of rice. The micro-array data has been deposited in the NCBI dbSNP Database under the accession codes 469281739 to 469324700.

Target probe preparation and 44 K SNP array hybridization. Rice genomic DNA was extracted from young green leaf tissue following Qiagen plantDNeasy protocol . Th e probe was generated using the BioPrime DNA labeling kit ( Invitrogen , Cat. No: 18094-011 ), and hybridization conditions were based on the Aymetrix SNP 6.0 protocol. Approximately 750 g to 1g of rice genomic DNA was labelled overnight at 25 C using 3 vol of the BioPrime DNA labelling reactions. Th e labelled DNA was ethanol precipitated, resuspended in 40 l H2O,

and then added to the Aymetrix SNP 6.0 hybridization cocktail. We did not include Human Cot-1 DNA because of the small size of the rice genome and the fact that it has a much smaller proportion of repetitive DNA compared with human or other mammals for which the assay was originally optimized.

SNP genotype calling . Genotypes are called using our program ALCHEMY, which was designed to provide improved performance in small sample sizes and for inbred populations with very low levels of heterozygosity 60. SNPs with low quality (that is, low call rate and allele frequency) across all samples were removed from the dataset and 36,901 high-performing SNPs (call rate > 70 % , minor allele frequency > 0.01) were used for all analyses. Of these SNPs, inbred samples had a median call rate of 95.9 % and pairwise concordances between technical replicates yielded>99% average pairwise concordance and>92% average call rate.

Plant materials. Th e Rice Diversity Panel consists of 413 Asian rice ( O. sativa) cultivars, including many landraces, which originated from 82 countries, representing all the major rice-growing regions of the world 15. Th e panel contains 87 indica, 57 aus, 96 temperate japonica, 97 tropical japonica, 14 groupV/aromatic, and 62 highly admixed accessions. All accessions were puried for two generations (single seed descent) before DNA extraction. In all, 20 of these 413 accessions were puried as part of the OryzaSNP project 6. Six cultivars (Azucena, Moroberekan, Nipponbare, Dom-Sod, IR64, M-202) were puried separately, once by Ali et al.15

and once as part of the OryzaSNP panel. Further information for each accession (accession name, accession number, country of origin and subpopulation ancestry based on PCA) is given in Supplementary Data 1 .

Phenotypic evaluation and correlation among individuals. Rice accessions were evaluated in the eld at Stuttgart, Arkansas during the growing season(May October) in 2006 and 2007. Two replications per year were grown in a randomized complete block design in single-row plots of 5 m length with a spacing of 25 cm between the plants and 0.50 m between the rows. A brief descriptionof each trait, its acronym, and evaluation methodology are summarized in

Supplementary Table S1 . Phenotypic correlations between individuals were calculated based on all phenotypes used in our study.

Estimation of LD decay in rice. Th e amount of genomic variation tagged by our SNP array was calculated by measuring the pairwise SNP linkage disequilibrium (LD) among the 44 K common SNPs (with MAF > 0.05) using r2, the correlation in frequency among pairs of alleles across a pair of markers. For all pairs of autosomal SNPs, r2 was calculated using the --r2 --ld-window 99999 --ld-window-r2 0 command in PLINK 61. Of the more than 44,100 SNP variants we assayed, we found 34,454 (~78%) with minor allele frequency >0.05 across the

O. sativa panel. When calculated across the entire O. sativa panel, LD is small at short distances ( r2<0.45 at 5 kb) but then decays more slowly, and still shows substantial residual LD at a distance of 2 Mb, reecting the deep subpopulation structure ( SupplementaryFig. S2 ). Within each subpopulation, we calculated r2 between all pairs of SNPs where both SNPs had < 20 % missing data and MAF 5%.

Population structure . Principal component analysis was done using EIGEN-SOFT 14. PC1 separates the samples into two main subspecies- Indica and Japonica and explains 34 % of the genetic variance whereas PC2 separates indica from aus and explains 10 % of the variance. We nd that PC3 separates the two japonica

groups into temperate and tropical components ( ~ 6 % of the variance), andPC4 identies the aromatic group as a clear and distinct gene pool ( ~ 2 % of the variance). (1- IBS) values were used as the distance between individuals to construct the hierarchical clustering tree using complete linkage method in Figure 2a .

Genome-wide association . Association analyses were performed with and without correcting for population structure. A mixed model approach implementedin EMMA 22 was used to correct the confounding of population structure. The relatedness matrix, measured as the genetic similarity between individuals andIBS values (that is, proportion of times a given pair of accessions had the same genotype across all SNPs), was used to estimate random eects. For all samples, SNPs and the top four PCs were used as xed eects; for association analysis within each subpopulation, only SNPs were used as xed eects in the model. For analyses without confounding, simple linear regression or logistic regression was used for continuous and binary traits, respectively. All statistical model details are described in the Supplementary Method . Unless explicitly mentioned, when two-year data were available, mean values across replicates and years of phenotypes were used in association analysis throughout the paper. To examine the eect of year on GWAS results, we introduced year as a covariate in the mixed model, along with the SNPs and 4 PCs. We graphed the correlation between P-values using the two-year-phenotypic mean and using year as a cofactor in the model for owering time and ag leaf length ( Supplementary Fig. S41 ). When examining GxE eects across locations, only 2007 owering time data from Arkansas was used for consistency with single-year data from the other locations. Candidate genes near hits were extracted from the literature. Rice homologues of Arabidopsis owering time genes were extracted from the Gramene Database ( www.gramene.org ).

Phenotypic variance contribution of signicant loci. To obtain signicant loci from EMMA for each phenotype, all signicant SNPs within 200 Kb were consolidated into one lowest P-value SNP to remove linkage disequilibrium. Large LD regions such as Hd1 were also consolidated into one single, most signicant SNP. Only continuous traits were considered for variance contribution estimation. SNP contribution to the phenotypic variance was estimated using ANOVA with the R package; statistical model details are provided in the Supplementary Method.

References

1. Toriyama, K., Heong, K. L. & Hardy, B. Rice is Life: Scientic Perspectives for the 21st Century: Proceedings of the World Rice Research Conference, Tsukuba, Japan (International rice research institute, 2005).

2. Greenland, D. J. Th e Sustainability of Rice Farming (Cab International, 1997).3. Go, S. A.et al. A dra sequence of the rice genome ( Oryza sativa L. ssp. japonica).Science 296, 92100 (2002).

4. Yu, J. et al. A dra sequence of the rice genome ( Oryza sativa L. ssp. indica). Science 296, 7992 (2002).

5. Huang, X. et al. Genome-wide association studies of 14 agronomic traits in rice landraces.Nat. Genet. 42, 961967 (2010).

6. McNally, K. L. et al. Genomewide SNP variation reveals relationships among landraces and modern varieties of rice . Proc. Natl Acad. Sci. USA 106, 1227312278 (2009).

7. Ebana, K. et al. Genetic structure revealed by a whole-genome single-nucleotide polymorphism survey of 5 diverse accessions of cultivated Asian rice ( Oryza sativa L.).Breeding Sci. 60, 390397 (2010).

8. Agrama, H. A., Yan, W., Jia, M., Fjellstrom, R. & McClung, A. M. Genetic structure associated with diversity and geographic distribution in the USDA rice world collection . Nat. Sci. 2, 247291 (2010).

9. Ebana, K., Kojima, Y., Fukuoka, S., Nagamine, T. & Kawase, M. Development of mini core collection of Japanese rice landrace . Breeding Sci. 58, 281291 (2008).

10. Project, R. A. Th e Rice Annotation Project Database (RAP-DB): 2008 update . Nucleic Acids Res. 36, D1028D1033 (2008).

11. Youens-Clark, K. et al. Gramene database in 2010: updates and extensions. Nucleic Acids Res. 39, D1085D1094 (2011).

12. Yamamoto, T., Yonemaru, J. & Yano, M. Towards the understanding of complex traits in rice: substantially or supercially?DNA Res. 16, 141154 (2009).

13. IRGSP. Th e map-based sequence of the rice genome . Nature 436, 793800 (2005).14. Price, A. L. et al. Principal components analysis corrects for stratication in genome-wide association studies.Nat. Genet. 38, 904909 (2006).15.Ali, M. L., McClung, A. M., Jia, M. H., Kimball, J. A.,McCouch, S. R. & Eizenga, G. C. A rice diversity panel evaluated for genetic and agro-morphological diversity between subpopulations and its geopgraphic distribution.Crop Sci. 51, doi:10.2135/cropsci2010.00.0641 (2011).

16. Zhao, K. et al. Genomic diversity and introgression in

O. sativa reveal the impact of domestication and breeding on the rice genome . PLOS One 5, e10780 (2010). 17. Garris, A. J., Tai, T. H., Coburn, J., Kresovich, S. & McCouch, S. Genetic structure and diversity in Oryza sativa L.Genetics 169, 16311638 (2005).

18. Takano-Kai, N. et al. Evolutionary history of

GS3, a gene conferring grain length in rice . Genetics 182, 13231334 (2009).19.Kovach, M. J., Calingacion, M. N., Fitzgerald, M. A. & McCouch,S. R. The origin and evolution of fragrance in rice ( Oryza sativa L.).Proc. Natl Acad. Sci. USA 106, 1444414449 (2009).

NATURE COMMUNICATIONS | 2:467 | DOI: 10.1038/ncomms1467 | www.nature.com/naturecommunications

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms1467

20.Fitzgerald, M. A., McCouch, S. R. & Hall, R. D. Not just a grain of rice: the quest for quality.Trends Plant Sci. 14, 133139 (2009).

21. Tian, Z. et al. Allelic diversities in rice starch biosynthesis lead to a diverse array of rice eating and cooking qualities . Proc. Natl Acad. Sci. USA 106, 2176021765 (2009).

22. Yu, J. et al. A unied mixed-model method for association mapping that accounts for multiple levels of relatedness . Nat. Genet. 38, 203208 (2006).

23. Kang, H. et al. Effi cient control of population structure in model organism association mapping.Genetics 178, 1709 (2008).

24. Zhao, K. et al. An Arabidopsis example of association mapping in structured samples.Plos Genet. 3, e4 (2007).

25.Sakamoto, T. & Matsuoka, M. Generating high-yielding varieties by genetic manipulation of plant architecture . Curr. Opin. Biotechnol. 15, 144147 (2004).

26.Xing, Y. & Zhang, Q. Genetic and molecular bases of rice yield.Annu. Rev. Plant Biol. 61, 421442 (2010).

27. Li, D. et al. Engineering OsBAK1 gene as a molecular tool to improve rice architecture for high yield . Plant Biotechnol. J. 7, 791806 (2009).

28. Komorisono, M. et al. Analysis of the rice mutant dwarf and gladius leaf 1. Aberrant katanin-mediated microtubule organization causes up-regulation of gibberellin biosynthetic genes independently of gibberellin signaling . Plant Physiol. 138, 19821993 (2005).

29. Famoso, A. N. et al. Genetic architecture of aluminum tolerance in rice( O. sativa) determined through genome-wide association analysis and QTL mapping.PLoS Genet. 7, e1002221 (2011).

30. Takeda, T. et al. The OsTB1 gene negatively regulates lateral branching in rice . Plant J. 33, 513520 (2003).

31. Ikeda, A. et al. slender rice, a constitutive gibberellin response mutant, is caused by a null mutation of the SLR1 gene, an ortholog of the height-regulating gene GAI/RGA/RHT/D8. Plant Cell 13, 9991010 (2001).

32. Yamamuro, C. et al. Loss of function of a rice brassinosteroid insensitive1 homolog prevents internode elongation and bending of the lamina joint . Plant Cell 12, 15911606 (2000).

33. Komatsu, M., Chujo, A., Nagato, Y., Shimamoto, K. & Kyozuka, J. FRIZZY PANICLE is required to prevent the formation of axillary meristems and to establish oral meristem identity in rice spikelets . Development 130, 38413850 (2003).

34. Asano, K. et al. SSD1, which encodes a plant-specic novel protein, controls plant elongation by regulating cell division in rice . Proc. Jpn Acad. Ser. B 86, 265273 (2010).

35. Wang, L. et al. OsLIC, a novel CCCH-type zinc nger protein with transcription activation, mediates rice architecture via brassinosteroids signaling.PLOS One 3, e3521 (2008).

36. Li, X. et al. Control of tillering in rice.Nature 422, 618621 (2003).37. McCouch, S. et al. Th rough the genetic bottleneck: O. rupogon as a source of trait-enhancing alleles for O. sativa . Euphytica 154, 317339 (2007).

38. Rieseberg, L. H., Widmer, A., Arntz, A. M. & Burke, B. Th e genetic architecture necessary for transgressive segregation is common in both natural and domesticated populations.Phil. Trans. R. Soc. Lond. B 358, 11411147 (2003).

39. Yano, M. et al. Hd1, a major photoperiod sensitivity quantitative trait locus in rice, is closely related to the Arabidopsis owering time gene CONSTANS. Plant Cell 12, 24732483 (2000).

40. Atwell, S. et al. Genome-wide association study of 107 phenotypes in Arabidopsis thaliana inbred lines.Nature 465, 627631 (2010).

41. Fujino, K. et al. Multiple introgression events surrounding the

Hd1 owering-time gene in cultivated rice, Oryza sativa L.Mol. Genet. Genom. 284, 137146 (2010).42. Hall, A. et al. The TIME FOR COFFEE gene maintains the amplitude and timing of arabidopsis circadian clocks . Plant Cell 15, 27192729 (2003).

43. Luesse, D. R., DeBlasio, S. L. & Hangarter, R. P. Plastid movement impaired 2, a new gene involved in normal blue-light-induced chloroplast movements in arabidopsis.Plant Physiol. 141, 13281337 (2006).

44. Takahashi, Y., Teshima, K. M., Yokoi, S., Innan, H. & Shimamoto, K. Variations in HD1 proteins, HD3A promoters, and EHD1 expression levels contribute to diversity of owering time in cultivated rice . Proc. Natl Acad. Sci. USA 106, 45554560 (2009).

45. Brachi, B. et al. Linkage and association mapping of Arabidopsis thaliana owering time in nature . PLoS Genet. 6, e1000940 (2010).

46.Bergelson, J. & Roux, F. Towards identifying genes underlying ecologically relevant traits in Arabidopsis thaliana . Nat. Rev. Genet. 11, 867879 (2010).

47. Yokoo, M., Kikuchi, F., Nakane, A. & Fujimaki, H. Genetic analysis of heading date by aid of close linkage with blast resistance in rice . Bull. Nat. Inst. Agric. Sci. Ser. D 31, 95126 (1980).

48. Fukuoka, S. et al. Loss of function of a proline-containing protein confers durable disease resistance in rice . Science 325, 9981001 (2009).

49. Liu, W.-q., Fan, Y.-y., Chen, J., Shi, Y.-f. & Wu, J.-l. Avoidance of linkage drag between blast resistance gene and the QTL conditioning spikelet fertility

based on genotype selection against heading date in rice . Rice Sci. 16, 2126 (2009).50. Nicolae, D. L. et al. Trait-associated SNPs are more likely to be eQTLs: annotation to enhance discovery from GWAS . Plos Genet. 6, e1000888 (2010).

51.Baker, M. Genomics: the search for association.Nature 467, 11351138 (2010).

52.Wang, C., Chen, S. & Yu, S. Functional markers developed from multiple loci in GS3 for ne marker-assisted selection of grain length in rice . Theor. Appl. Genet. 122, 905913 (2011).

53.Kovach, M. J., Sweeney, M. T. & McCouch, S. R. New insights into the history of rice domestication.Trends Genet. 23, 578587 (2007).

54.Brown, J. K. M. Yield penalties of disease resistance in crops.Curr. Opin. Plant Biol. 5, 339344 (2002).

55.Boerma, H. R. & Walker, D. R. Discovery and utilization of QTLs for insect resistance in soybean.Genetica 123, 181189 (2005).

56.Mitchell-Olds, T. Complex-trait analysis in plants.Genome Biol. 11, 113 (2010).57. Tung, C.- W. et al. Development of a research platform for dissecting phenotype genotype associations in rice ( Oryza spp.).Rice 3, 205217 (2010).

58. McCouch, S. R. et al. Development of genome-wide SNP assays for rice. Breeding Sci. 60, 524535 (2010).

59.Ammiraju, J. S. S.et al. The

Oryza bacterial articial chromosome library resource: construction and analysis of 12 deep-coverage large-insert BAC libraries that represent the 10 genome types of the genus Oryza. Genome Res. 16, 140147 (2006).60. Wright, M. H. et al. ALCHEMY: a reliable method for automated SNP genotype calling for small batch sizes and highly homozygous populations . Bioinformatics 26, 29522960 (2010).

61. Purcell, S. et al. PLINK: a tool set for whole-genome association and population-based linkage analyses.Am. J. Hum. Genet. 81, 559575 (2007).

Acknowledgements

We thank Teresa Hancock and Daniel Wood at the University of Arkansas Rice Research and Extension Center, Stuttgart, Arkansas for their outstanding technical assistance; Shoqul Islam at Bangladesh Agricultural University for managing the eld experiments; Peter Schweitzer, Wei Wang and Barbara Hover from the Cornell Genomics Facility for excellent technical support; Robert Barkovich, Julia Montgomery, Gene Tanimoto and Ali Pirani from Aymetrix for advise designing the 44 K chip; OryzaSNP project for early access to SNP data; Joshua Cobb for help with candidate gene annotation; Ellie Rice,Dan Deibler, Cheryl Utter and Shelina Gautama for images design; and Simon Gravelfor valuable discussion during manuscript preparation. We are grateful for generous computing support from USC CEGS (NIH CEGS grant P50 HG002790; S. Tavar , PI) and Stanford Biox2 clusters. The owering-time data from Aberdeen and Faridpurare outputs from grant BBF0041841 funded by BBSRC-DFID (UK) awarded to Andy Meharg (Aberdeen) and AP. Th e development of the 44 K SNP array, rice diversity panel, genotyping dataset and phenotypic evaluation of 32 traits was supported by NSF Plant Genome Research Program award # 0606461 to S.R.M., G.C.E., A.M.M. and C.D.B.

Author contributions

K.Z. and C.-W.T. contributed equally to the work, and C.D.B. and S.R.M. co-supervised the project. K.Z., C.-W.T., S.R.M., C.D.B., G.C.E. and A.M.M. conceived and designed the experiments. . K.Z., C.-W.T., M.H.W., G.C.E., M.L.A., G.J.N., R.I., A.M.M. and A.H.P. performed the experiments. K.Z., C.-W.T., M.H.W., A.R. and S.R.M.,analysed the data. K.Z., C.-W.T., M.H.W., G.C.E., A.M.M., J.M. and S.R.M. contributed reagents / materials /

analysis tools. K.Z., C.-W.T., J.M., C.D.B., S.R.M. wrote the paper.

Additional information

Accession codes: Th e microarray data has been deposited in the NCBI dbSNP Database under the accession codes 469281739 to 469324700.

Supplementary Information accompanies this paper at http://www.nature.com/ naturecommunications

Competing nancial interests: Th e authors declare no competing nancial interests.

Reprints and permission information is available online at http://npg.nature.com/ reprintsandpermissions/

How to cite this article: Zhao, K. et al. Genome-wide association mapping reveals a rich genetic architecture of complex traits in Oryza sativa. Nat. Commun. 2:467 doi: 10.1038 / ncomms1467 (2011).

License: Th is work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivative Works 3.0 Unported License. To view a copy of this license, visit http:// creativecommons.org/licenses/by-nc-nd/3.0/

NATURE COMMUNICATIONS | 2:467 | DOI: 10.1038/ncomms1467 | www.nature.com/naturecommunications

Word count: 9754

Show less

Abstract

Translate

Asian rice, Oryza sativa is a cultivated, inbreeding species that feeds over half of the world's population. Understanding the genetic basis of diverse physiological, developmental, and morphological traits provides the basis for improving yield, quality and sustainability of rice. Here we show the results of a genome-wide association study based on genotyping 44,100 SNP variants across 413 diverse accessions of O. sativa collected from 82 countries that were systematically phenotyped for 34 traits. Using cross-population-based mapping strategies, we identified dozens of common variants influencing numerous complex traits. Significant heterogeneity was observed in the genetic architecture associated with subpopulation structure and response to environment. This work establishes an open-source translational research platform for genome-wide association studies in rice that directly links molecular variation in genes and metabolic pathways with the germplasm resources needed to accelerate varietal development and crop improvement.

Details

Title

Genome-wide association mapping reveals a rich genetic architecture of complex traits in Oryza sativa

Author

Zhao, Keyan; Tung, Chih-wei; Eizenga, Georgia C; Wright, Mark H; Ali, M Liakat; Price, Adam H; Norton, Gareth J; Islam, M Rafiqul; Reynolds, Andy; Mezey, Jason; Mcclung, Anna M; Bustamante, Carlos D; Mccouch, Susan R

Pages

467

Publication year

2011

Publication date

Sep 2011

Publisher

Nature Publishing Group

e-ISSN

20411723

Source type

Scholarly Journal

Language of publication

English

DOI

https://doi.org/10.1038/ncomms1467

ProQuest document ID

894311696

Genome-wide association mapping reveals a rich genetic architecture of complex traits in Oryza sativa

Jump to:

Full text

Abstract

Details

Suggested sources