GWAS provide genetic and biochemical insights into natural variation in rice metabolism

Articles

Genome-wide association analyses provide genetic and biochemical insights into natural variation in rice metabolismWei Chen1,4, Yanqiang Gao1,4, Weibo Xie1,4, Liang Gong1,4, Kai Lu1,4, Wensheng Wang1, Yang Li1, Xianqing Liu2, Hongyan Zhang3, Huaxia Dong1, Wan Zhang1, Lejing Zhang1, Sibin Yu1, Gongwei Wang1, Xingming Lian1 & Jie Luo1Plant??metabolites??are??important??to??world??food??security??in??terms??of??maintaining??sustainable??yield??and??providing??food??with??enriched??phytonutrients.??Here??we??report??comprehensive??profiling??of??840??metabolites??and??a??further??metabolic??genome-wide??association??study??based??on??~6.4??million??SNPs??obtained??from??529??diverse??accessions??of??Oryza sativa.??We??identified??hundreds??of??common??variants??influencing??numerous??secondary??metabolites??with??large??effects??at??high??resolution.??We??observed??substantial??heterogeneity??in??the??natural??variation??of??metabolites??and??their??underlying??genetic??architectures??among??different??subspecies??of??rice.??Data??mining??identified??36??candidate??genes??modulating??levels??of??metabolites??that??are??of??potential??physiological??and??nutritional??importance.????As??a??proof??of??concept,??we??functionally??identified??or??annotated??five??candidate??genes??influencing??metabolic??traits.??Our??study??provides??insights??into??the??genetic??and??biochemical??bases??of??rice??metabolome??variation??and??can??be??used??as??a??powerful??complementary??tool??to??classical??phenotypic??trait??mapping??for??rice??improvement.

Plants produce a vast array of chemically and biologically different compounds. Plant metabolites are not only important for plants them-selves and their interactions with the environment1,2 but also provide indispensable resources for humans as sources of nutrition, energy and medicine1,3,4. For example, vitamins such as the water-soluble B6 group have been reported to reduce the incidence of important human diseases such as hypertension and diabetes5. Recent reports also demonstrated their function in cellular defense against oxida-tive stress in plants6. In addition, secondary metabolites such as phenolamides and flavonoids have indispensable roles in chemical defenses against biotic and abiotic stresses7,8, and flavonoids also confer health-promoting effects against chronic diseases and certain cancers in humans9–11. The extreme quantitative and qualitative vari-ations in metabolites have made plants the ideal models for dissecting the biosynthetic pathways and regulation of metabolites12,13.Cultivated rice (Oryza sativa L.) is one of the most important crops and feeds approximately half of the human population of the world. Understanding the genetic basis of natural variation of the metabolome in major crops such as rice is essential for the qual-ity, reliability and sustainability of the world’s food supply. Rice landraces have evolved from their wild progenitor and show high genetic diversity14,15. Understanding the genetic and biochemical bases of metabolic traits among these diverse varieties will pro-vide important insights for breeding elite varieties with increased resistance to detrimental stresses and enhanced nutrition for humans. So far, study of the genetic architecture of metabolic traits in rice has been based on quantitative trait locus (QTL) linkage mapping using bi-parental populations16. Although providing valu-able insights17, this approach is clearly not scalable to investigate the tremendous variation of abundant diverse germplasm resources. Moreover, the traditional ways of dissecting biosynthetic pathways are usually slow and tedious4. Consequently, the identities, bio-synthetic pathways and regulation of thousands of metabolites are largely unknown13,18, as is the genetic basis for variation in metabolite levels.Genome-wide association study coupled with metabolomics analysis (or mGWAS) makes it possible to screen a very large number of accessions simultaneously to understand genetic contributions to metabolic diversity and their relevance to complex traits19–21. For example, mGWAS on both primary21 and secondary20 metabo-lites have been carried out in Arabidopsis thaliana with large-scale validation of new links between genes and metabolites, such as the linkages to glucosinolates, although the biochemical relevance of these gene-metabolite associations remains to be explored. Despite progress in model plants, few such studies have been carried out in crops and none have been done in rice, which is an ideal candidate system for GWAS because of its self-fertilization and high-quality 15,22reference genome.Here, by performing comprehensive metabolic profiling and an mGWAS, we identify complex and distinct genetic regulation of metabolites in two subspecies of rice. We also demonstrate that the mGWAS provides a powerful tool for large-scale interactive gene-metabolite annotation and identification, pathway elucidation and knowledge about crop improvement.p© 2014 Nature America, Inc. All rights reserved

.

2College of Life Science and Technology, Huazhong Agricultural University, Wuhan, China. 3Key Laboratory of Horticultural Plant Biology (Ministry of Education), Huazhong Agricultural University, Wuhan, China. 4These authors contributed equally to this work. Correspondence should be addressed to J.L. (jie.luo@http://wendang.chazidian.com) or X. Lian (xmlian@http://wendang.chazidian.com).1National Key Laboratory of Crop Genetic Improvement and National Center of Plant Gene Research (Wuhan), Huazhong Agricultural University, Wuhan, China.

Received 18 December 2013; accepted 15 May 2014; published online 8 June 2014; doi:10.1038/ng.3007

Nature GeNetics

Articles

table 1 summary of genome-wide significant associations identified in the three GWAs panelsRESULTS

Metabolic??profiling??of??rice??leavesPopulationDetected in at least Detected in at least We determined the levels of 840 distinct met-Itemone populationtwo populationsAllindicajaponicaabolic traits in the leaves of rice plants at the Number of traits441392391598323five-leaf stage (termed ‘leaf’ hereafter) using Number of lead SNPsa1,4568971,2302,947551a newly developed liquid chromatography– Number of locib434293376634356tandem mass spectrometry (LC-MS/MS)-SNPs above 20% of variation5073366961,114420based, widely targeted metabolic profiling Maximum explained variation (%)89.986.183.889.989.9

19.421.223.919.729.6method (Online Methods and Supplementary Explained variation per SNP (%)Tables 1 and 2). Of these metabolites, 277 There are three independent P values, after Bonferroni correction, for significant associations in the three GWAS were identified or annotated (Supplementary panels: all (i.e., the full population) (6.6 × 10?8), indica (8.7 × 10?8) and japonica (2.0 × 10?7).

aThe SNPs with the lowest P value in a defined region. bAdjacent lead SNPs separated by less than 300 kb were considered as Table 3). Among the 840 metabolites deter-a cluster.mined in the diverse global collection of O. sativa accessions (Supplementary Tables 4 and 5), 58.7% displayed broad-sense heritability (H2) greater than (corresponding to 356 loci) were detected repeatedly (for example, 0.5, and 24.4% had heritability over 0.7 (Supplementary Fig. 1a). For in at least two populations) (Supplementary Tables 8–11). 71.2% the majority of the metabolites, the observed coefficients of variation of the metabolites detected (598 out of 840) had at least one signifi-cant association, with an average of 4.9 associations per metabolite. were greater than 50% (Supplementary Fig. 1b).The population structure of the worldwide rice germplasm col-In general, these loci show large effects to explain the variation: up lection has been well characterized23–25. However, little is known to 90%, with an average of 19.7% (Table 1). The significant asso-about how these distinctions affect small-molecule profiles26–29. On ciations are illustrated in Supplementary Figure 4. Given both the the basis of the levels of metabolites detected in our study, cluster false-positive and false-negative issues associated with GWAS21,32,33, analysis grouped the varieties into two distinct clusters: indica and the full lists of significant and suggestive associations are presented japonica (Supplementary Fig. 2a). Principal component analysis in Supplementary Tables 12 and 13, which could be used for further of all the metabolic data yielded similar results (Supplementary validation and follow-up study. Manhattan plots of the significant Fig. 2b). We selected metabolites with levels that varied more than loci detected repeatedly are also illustrated, including 161 loci corres-tenfold among different subspecies (subspecies differentiation metabo-ponding to flavonoids, phenolamides, amino acids and their deriva-lites) (Supplementary Fig. 3) and calculated the differences in their tives, terpenoids, nucleic acids and their derivatives and other known mean contents within the two subspecies (Supplementary Table 6). metabolites, as well as 195 loci corresponding to currently unknown Most of the metabolites that accumulated preferentially in the indica metabolites (Fig. 1 and Supplementary Table 14).Although the levels of some metabolites were controlled by only accessions (cluster 1) are C-glycosylated and malonylated flavonoids, whereas the japonica varieties had substantially higher levels of one major locus that explained over 50% of the natural variation, 30phenolamides and arabidopyl alcohol derivatives (cluster 2). such as mr1002 (sakuranetin), mr1072 (luteolin 6-C-glucoside) and These subspecies-specific metabolites likely reflect, as well as affect, mr1750, the levels of most metabolites were determined by multiple the subspecies differentiation of rice, especially between the two loci. According to the mGWAS results, 415 of the detected metabolites have multiple significant loci (Supplementary Table 9). To test for pos-major subspecies, indica and japonica.sible interactions between significant loci, we calculated the pairwise epistatic interactions between the significant loci against the average Genetic??basis??of??natural??variation??in??different??subspeciesWe sequenced a diverse global collection of 529 O. sativa accessions accumulation of 415 metabolites within the full population. A total of (Supplementary Table 4) using the Illumina HiSeq 2000 system, gen-3,351 significant interactions (P < 0.001) were detected for 241 metabo-erating a total of 6.7 billion 90-bp paired-end reads (NCBI BioProject lites (Supplementary Table 15). For example, we observed significant accession number PRJNA171289). We then added sequences from interaction between the two major associated loci underlying the level 14 950 varieties generated by Huang et al. to facilitate SNP identifica-of mr1134 (putative arabidopyl ketoadipic acid). The effect of the tion and genotype imputation. After imputation, we selected SNPs G allele at SNP sf0138154786 (with the SNP number indicating the locus, with missing rates of less than 20%, resulting in a total of 6,428,770 for example, sf0138154786 means on chromosome 1 at 138,154,786 bp) SNPs with a missing data rate of approximately 0.38% for further on increasing mr1134 content was dependent on the G allele at the analysis (Online Methods). The SNPs and imputed genotypes were SNP sf0100094775 locus (Fig. 2a), suggesting that these loci may act pre-released and are available for query on our website (RiceVarMap; sequentially in the biosynthesis of mr1134 (Supplementary Fig. 5a). We observed a different scenario for the interaction between SNP see URLs).We then performed mGWAS for both the full population (524 lines sf0707666727 (C or T) and sf0408290032 (G or A) in controlling from the sequencing panel) and each of the two subspecies of rice, mr1083 (putative tricin 5-O-hexoside) content. The varieties with a indica (295 lines) and japonica (156 lines), using both simple linear genotype of T and G at the two loci, respectively, contain no tricin 31regression (LR) and a linear mixed model (LMM). The LMM 5-O-hexoside. However, varieties with genotypes of other combi-resulted in fewer false positives by taking into account the genome-nations at these two loci accumulate either much higher or similar wide patterns of genetic relatedness15, and we used these results for amounts of tricin 5-O-hexoside (Fig. 2b). Therefore, it is possible that further analysis. We used PLMM = 6.6 × 10?8, 8.7 × 10?8 and 2.0 × 10?7 these two loci act in parallel to determine tricin 5-O-hexoside levels, as the genome-wide significance thresholds for the full population and where T and G are the nonfunctional alleles at SNPs sf0707666727 and the indica and japonica subpopulations, respectively, after Bonferroni sf0408290032, respectively (Supplementary Fig. 5b).Genome-wide analysis of the significant loci identified a signifi-correction (Supplementary Table 7 and Online Methods). We detected a total of 2,947 lead SNPs, corresponding to 634 cant deviation from random distribution across the 12 chromosomes 2?162?6loci in at least one of the populations, within which 551 lead SNPs (χ = 240.04, P < 2.2 × 10 for indica and χ = 48.08, P < 1.4 × 10 p© 2014 Nature America, Inc. All rights reserved.

Nature GeNetics

Articles

Figure 1 Manhattan plots of mGWAS results with genetic association. The strength of association for known (top) and unknown (bottom) metabolites is indicated as the

negative logarithm of the P value for the LMM model. All metabolite-SNP associations with P values below 6.6 × 10?8 (horizontal dashed lines in all Manhattan plots) are plotted against genome location in intervals of 1 Mb. Triangles represent metabolite-SNP associations with P values below 1.0 × 10?40. AA and NA ders, amino acid and nucleic acid derivatives.

4030–log10 P

20100010–log10 P

203040

Chr.

1

2

3

4

5

6

7

8

9

10

11

12

© 2014 Nature America, Inc. All rights reserved.

for japonica), suggesting that major genes Flavonoidscontrolling the levels of large sets of metabo-lites exist within regions enriched for the significant loci. We identi-fied 11 and 14 potential mGWAS hot spots in the indica and japonica subspecies, respectively. mGWAS hot spots in indica rice were located mainly on chromosomes 2, 6 and 12, but the hot spots were located on chromosomes 4 and 12 in japonica (Fig. 2c and Supplementary Table 16). When we compared loci for individual metabolites between the two subspecies, we detected 359 distinct loci among the 514 total loci detected in the two subspecies (Fig. 2d and Supplementary Table 17), suggesting that the majority of loci are under different genetic control in the different subspecies. Such differences were reflected in both the associated loci detected and their effect sizes. Notably, the genetic architecture was quite different between most of the subspecies dif-ferentiation metabolites (Fig. 2e and Supplementary Table 18). For example, we identified one major associated locus underlying luteolin 6-C-glucoside content in the indica panel, whereas we detected no significant association for the same metabolite in the japonica panel. Despite the overall subspecies-specific regulation of metabolism, we also found metabolites that were under the control of either modestly similar (mr1699) or the same genetic architecture (Fig. 2f).

Intensity

(c.p.s., × 104)

PhenolaminesTerpenoidsAA and NA dersOthers,Unknown

From these results, we conclude that naturally occurring variation in the rice leaf metabolome is under complex genetic control and distinct networks of genes regulate the leaf metabolome in different subspecies of rice.

Large-scale??metabolite??identification??and??annotation

Our mGWAS facilitated the identification or annotation of the metabolites tested by linking the unknown metabolites to functionally related genes (Supplementary Figs. 6–9). For example, the associa-tion between SNP sf1207801034 within NOMT, the gene encod-ing naringenin 7-O-methyltransferase34, and the level of mr1002 (m/z 287.1) suggested that this metabolite could be sakuranetin. We subsequently identified the mr1002 metabolite as sakuranetin by com-paring the retention time and fragmentation pattern of this metabolite with those of a commercial standard. We also observed significant

?7associations between the levels of mr1158 (P = 1.4 × 10) and mr1168

(P = 4.0 × 10?10) and the SNPs sf0700012531 and sf1000079974, respectively. Notably, these SNPs are intronic within putative genes encoding SOR/SNZ family proteins (Os07g01020 and Os10g01080, respectively) that both have high homology with the pyridoxine synthase gene At5G01410 (85% and 80% identity at the amino acid level, respectively). These data indicate that the metabolites associ-ated with these loci might be pyridoxine or one of its derivatives. We then identified mr1158 as pyridoxine by using an authentic standard and annotated mr1168 as pyridoxine O-glycoside. These identifications, in turn, suggested strongly that Os07g01020 and Os10g01080 encoded pyridoxine synthases. Similarly, the strong association between SNP sf0221807001, which is in linkage disequi-librium (LD) with a gene encoding cytochrome P450 monooxygenase (CYP76M7) that is involved in the production of phytocassanes35, and the level of mr1042 (m/z 317.2097) facilitated the identifica-tion of this metabolite as phytocassane D. The comparison of the exact mass and fragmentation pattern of mr1042 with those of

the

phytocassane D standard further support this annotation.

Figure 2 Two-locus interactions between significant loci and mGWAS

genetic architecture analysis between the indica and japonica subspecies. (a) Interaction between two major loci controlling mr1134 (putative arabidopyl ketoadipic acid) accumulation. c.p.s., counts per second. (b) Interaction between two major loci controlling mr1083 (putative tricin 5-O-hexoside) accumulation. Error bars (a,b), s.e.m. (c) Distribution of mGWAS signals across the rice genome in the indica and japonica subspecies. (d) Significant loci detected in the indica and japonica

subspecies. Shown are 359 distinct and 155 co-detected loci among the total 514 loci detected in the two subspecies. (e) Significant loci detected for subspecies differentiation metabolites in the indica and japonica subspecies. (f) Examples of significant loci detected for individual subspecies differentiation metabolites in the two subspecies of rice.

a

pc

Number of significant loci

indica

4020020100

japonica

12345indica

6789101112

d

indica138

155

japonica221

f

–log10 P

402002010

mr1072mr1404mr1699

e

indica72

15

japonica66

japonica

Chr.1

23456789101112

Nature GeNetics

Articles

table 2 summary of 36 candidate genes assigned from mGWAs results

MetaboliteSmiglaside C

Apigenin 7-O-glucosideN-sinapoylputrescine

Threonyl carbamoyl adenosineArabidopyl ketoadipic acidTricin O-malonylhexosideTricin O-malonylhexosidePhytocassane D

Luteolin 6-C-glucosideTrigonellineUnknownl-alanine

Kynurenic acid

Methylapigenin C-hexosideN-feruloylagmatineSyringenone

Trans-zeatin N-glucosideDIMBOA glucosideCyanidin 3-O-glucoside4-pyridoxic acid O-hexosideMethylnaringenin C-pentosideDi-C,C-pentosyl-apigeninPyridoxine

Inosine 5′-monophosphateApigenin 5-O-glucoside(?)-trans-carveolIntegrifoside A

N-feruloylputrescinePyridoxine O-glucosideChlorogenic acidl-tyramine

Chrysoeriol 7-O-rutinosideUnknown

p-coumaroyl-2-hydroxyputrescineSakuranetin

N-p-coumaroylspermidine

P9.2 × 10?477.4 × 10?112.1 × 10?521.0 × 10?261.7 × 10?262.9 × 10?252.9 × 10?251.3 × 10?241.0 × 10?412.1 × 10?366.9 × 10?892.3 × 10?73.6 × 10?212.2 × 10?434.1 × 10?151.0 × 10?903.0 × 10?151.9 × 10?81.0 × 10?252.7 × 10?253.5 × 10?651.7 × 10?521.4 × 10?73.6 × 10?154.6 × 10?331.3 × 10?143.6 × 10?466.3 × 10?224.0 × 10?101.1 × 10?157.8 × 10?85.3 × 10?113.4 × 10?207.0 × 10?594.6 × 10?286.9 × 10?98

Possible causative SNPsf0106733065asf0130712131sf0133590589sf0137865211sf0138138908sf0216673251sf0216745114sf0221734929sf0222736277sf0235364705bsf0314570065csf0314775019sf0405640502sf0406560429sf0433744525sf0506883423sf0507153510sf0523814956vf0605315059dsf0609997194sf0610554080sf0610587603sf0700011045sf0716108195sf0719060887sf0800288026sf0802210305sf0921463527sf1000079974sf1006593981sf1012189820sf1115033398sf1125035756sf1125117890sf1207801034sf1215967910a

Candidate geneOs01g12330Os01g53460Os01g58070Os01g65260Os01g65680Os02g28170Os02g28340Os02g36110Os02g37690Os02g57760Os03g25500Os03g25820Os04g10410Os04g11970Os04g56910Os05g12040Os05g12450Os05g40720Os06g10350Os06g17260Os06g18140Os06g18670Os07g01020Os07g27580Os07g32060Os08g01450Os08g04500Os09g37200Os10g01080Os10g11860Os10g23900Os11g26950Os11g42370Os11g42480Os12g13800Os12g27220

DescriptionExpressed proteinUGT

Polyphenol oxidaseAPB transferaseDOPA dioxygenaseTransferaseTransferase

Cytochrome P450UGT

O-methyltransferaseCytochrome P450Amino acid permeaseAmidase

O-methyltransferaseTransferase

Cytochrome P450Hydroquinone UGTUGTOsC1UGTUGTUGT

SOR/SNZ proteinHAD phosphataseUGT

Cytochrome P450Terpene synthaseTransferase

SOR/SNZ proteinMATE efflux proteinDecarboxylaseUGT

TransferaseTransferase

O-methyltransferaseTransferase

Function identification/annotation

ref. 36

ref. 16ref. 16ref. 36In vitro/in vivo

In vivo

© 2014 Nature America, Inc. All rights reserved.

ref. 37

In vivoref. 38

In vitro/in vivo

ref. 16In vivoref. 34

aSNP introducing stop codon. bSNP introducing initiation codon. cSNP disrupting stop codon. d10-bp deletion. More information is listed in supplementary table 20.

pIn total, the GWAS enabled the identification and/or annotation of a total of 166 metabolites (Supplementary Table 19).

Biochemical??and??functional??interpretation??of??GWAS??resultsBesides providing insights into the genetic basis of metabolome vari-ation, an mGWAS also provides possible biochemical and functional insights into the underlying pathways. Association signals were located close to or landed directly on eight genes that have been reported pre-viously16,34–38, demonstrating the relatively high resolution of our mGWAS. We then mined candidate genes that have not been identi-fied previously by (i) looking for a protein or protein cluster that is biochemically related to the associated metabolic trait encoded at these loci; (ii) performing cluster analysis of candidate genes relative to homologous genes with known function; and (iii) crossreferencing with results from linkage mapping. By applying these approaches, we were able to identify a number of candidate genes and possible causa-tive SNPs underlying metabolites that are of potential plant physiolog-ical and/or human nutritional importance (Table 2, Supplementary Fig. 10 and Supplementary Table 20). The Manhattan plots of these SNPs are shown in Supplementary Figure 11.

Trigonelline, the N-methyl conjugate of nicotinic acid, has long been reported to regulate various processes, especially abiotic

Nature GeNeticsstresses in plants39,40, but the enzyme that catalyzes the key step in the formation of trigonelline from nicotinic acid is unknown. Levels of trigonelline were significantly associated (P = 2.1 × 10?36) with SNP sf0235317720 on chromosome 2 that is in LD with Os02g57760, which encodes a protein annotated as an O-methyltransferase, suggesting that Os02g57760 encodes a methyltransferase that catalyzes the key step of trigonelline biosynthesis (Fig. 3 and Supplementary Fig. 10a).

SNP sf1012174133, on chromosome 10, is located 15 kb from Os10g23900 (encoding a putative decarboxylase), and this SNP was significantly associated (P = 7.8 × 10?8) with l-tyramine. The high sequence identity (43% at the amino acid level) between Os10g23900 and AtTyrDC41, an l-tyramine decarboxylase in Arabidopsis, sug-gests that Os10g23900 is likely the candidate gene underlying this locus. Similarly, Os03g25820 (encoding a putative amino acid per-mease) was assigned as the candidate gene underlying the levels of l-alanine and some other amino acids, such as l-threonine, l-leucine and l-histidine.

Clear signals for phenolamides also resulted in the assignment of candidate genes for these subspecies differentiation compounds. The significant association (P = 6.3 × 10?22) between SNP sf0921455575 that lies 8 kb upstream of Os09g37200 (encoding a putative transferase)

Articles

Figure 3 Functional identification of Os02g57760 (O-methyltransferase) and the assignment of possible causative sites. (a) Structure and

LC-MS/MS fragmentation of trigonelline. (b) Manhattan plot displaying the GWAS result of the content of trigonelline. (c) Gene model of

Os02g57760. Filled black boxes represent coding sequence. The gray vertical lines mark the polymorphic sites identified by high-throughput sequencing, and the star represents the proposed functional site. (d) A representation of pairwise r2 values (a measure of LD) among all polymorphic sites in Os02g57760, where the darkness of the color of each box corresponds to the r2 value according to the legend. (e) Box plot (the middle line indicates the median, the box indicates the range of the 25th to 75th percentiles of the total data, the whiskers indicate the interquartile range and the outer dots are outliers) for trigonelline content, plotted as a function of genotypes at SNP sf0235364705. The metabolic data of trigonelline were log2 transformed. (f) LC-MS chromatograms of in vitro enzyme assays showing the enzyme activity of recombinant Os02g57760. Protein extract from E. coli containing the pDEST17 empty vector was used as a negative control. (g) In vivo function of Os02g57760. Shown are bar plots for the mRNA level of Os02g57760 (left) and for the content of trigonelline (middle) and nicotinic acid (right) in transgenic positive individuals. WT, the transgenic background variety ZH11. Data are shown as the means ± s.e.m., n = 3.

a

Intensity (c.p.s., × 106)

e

Relative content

201050

21181540302010

Intensity (c.p.s., × 10)

4

Trigonelline

Standards

b

30

–log10 P

f

Nicotinic acid

2010096320

Empty vector

cd

550858

Os02g57760

3

specifically with the levels of different N-hydroxycinnamoyl- putrescine compounds suggests that Os09g37200 encodes a putrescine hydroxycinnamoyl acyltransferase (Supplementary Fig. 12). Similarly, the strong association (P = 4.1 × 10?15) between SNP sf0433733272 at the Os04g56910 locus and the level of N-hydroxycinnamoyl agmatine suggests that Os04g56910 encodes an agmatine hydroxycinnamoyl acyltransferase (Supplementary Fig. 13 and Supplementary Table 21). Os12g27220 was tentatively assigned as a spermidine acyltransferase, with SNP sf1215974303 being the candidate polymorphism for the natural variation in levels of N-hydroxycinnamoyl spermidines in rice (Supplementary Fig. 10b).CGA (chlorogenic acid) and flavonoids act as antioxidants in plants and protect against degenerative, age-related diseases in humans when supplied in the diet10. We found a strong association (P = 1.1 × 10?15) between CGA content and SNP sf1006572031 that was located 21 kb upstream of Os10g11860. Cluster analysis and the high sequence identity between Os10g11860 and AtTT12 (36% at the amino acid level), the MATE efflux protein responsible for the transport of proanthocyanin42, strongly suggest Os10g11860 as a can-didate gene underlying the metabolic QTL (mQTL) (Supplementary Fig. 10c). SNP sf0719048913, located 11 kb away from Os07g32060 (encoding a putative UDP-glucosyl transferase), was significantly

?33associated (P = 4.6 × 10) with a number of flavone 5-O-glucosides,

suggesting that Os07g32060 encodes a flavone 5-O-glucosyltransferase. This association is also supported by results from linkage mapping and phylogenetic analysis (Supplementary Fig. 10d). Further analysis indicated that the exonic SNPs 8–14 (in high LD with each other; 2?47r = 0.94–0.99) were significantly associated (P = 7.61 × 10 to 1.07 × 10?45) with the target trait and could be functional polymorphisms underlying this GWAS hit (Fig. 4 and Supplementary Table 21). Similarly, we could assign Os11g26950 as a candidate gene underlying flavonoid O-rutinoside content (Table 2 and Supplementary Fig. 10d). Of the 17 loci that were significantly associated with levels of C-glycosyl flavones—one of the major groups of metabolites differ-entiating subspecies of rice—we were able to assign four candidate genes. In particular, Os06g18140 was tentatively annotated as a flavone C-pentosyltransferase, Os06g18670 as a C-pentosyl flavone pentosyl-transferase, Os02g37690 as a flavone/flavanone C-hexosyltransferase and Os04g11970 as a flavonoid methyltransferase on the basis of their specific associations with certain classes of metabolites (Table 2 and Supplementary Fig. 10a,d).

© 2014 Nature America, Inc. All rights reserved.

Relative expression

Os02g5776080400

WT4

51215

Relative content (× 106)

Relative content (× 10)

g

83204132232351

222303

1025

508220480

2

101

12345

Retention time (min)543210

WT451215

Nicotinic acid

6

Trigonelline12840

WT4

51215

Functional??identification??or??annotation??of??candidate??genesAlthough validation of all associations disclosed is beyond the scope of a single study, we nevertheless attempted to show that, in principle, such validation is possible. To this end, we experimentally verified the assignment and further characterized some of candidate genes, providing new biochemical and functional insights (Table 2 and Supplementary Table 20).

The association of Os02g57760 with trigonelline suggests that Os02g57760 is the methyltransferase that catalyzes the key step of trigonelline biosynthesis. When Os02g57760 was expressed with a histidine tag at the N-terminus in Escherichia coli BL-21, we were able to detect nicotinic acid:N-methyltransferase activity in the soluble protein extract (Fig. 3f). Consistent with its in vitro activity, over-expression of Os02g57760 in ZH11, a japonica variety that exhibits low trigonelline content, resulted in a substantial increase in trigonelline accumulation in the transgenic-positive lines compared to control plants, confirming its function in vivo (Fig. 3g). Similarly, Os09g37200 was characterized as a putrescine/agmatine hydroxycinnamoyl acyltransferase by in vitro enzyme assay of recombinant Os09g37200 protein and subsequent transgenic analysis (Supplementary Fig. 12 and Supplementary Table 21).

The functional annotations of candidate genes that have not been characterized fully were further supported by showing that over-expression of these genes resulted in specific increases in the corres-ponding metabolites. For example, we assigned Os07g32060 as the candidate gene underlying the QTL for flavone 5-O-glucosides and annotated this gene as flavone 5-O-glucosyltranferase (Table 2 and Fig. 4). When this gene was overexpressed in rice, we detected up

Nature GeNetics

p4