Molecular Plant • Volume 5 • Number 2 • Pages 442–451 • March 2012 RESEARCH ARTICLE

Deciphering Starch Quality of Rice Kernels

Using Metabolite Profiling and Pedigree
Network Analysis
Miyako Kusanoa,b,1, Atsushi Fukushimaa, Naoko Fujitac, Yozo Okazakia, Makoto Kobayashia,
Naoko Fujita Oitomec, Kaworu Ebanad and Kazuki Saitoa,e
a RIKEN Plant Science Center, Tsurumi, Yokohama 230-0045, Japan
b KIHARA Institute for Biological Research, Yokohama City University, Totsuka, Yokohama 244-0813, Japan
c Department of Biological Production, Akita Prefectural University, Akita 010-0195, Japan
d National Institute of Agrobiological Sciences, 2–1–2 Kannondai, Tsukuba, Ibaraki 305-8602, Japan
e Graduate School of Pharmaceutical Sciences, Chiba University, Chiba 260-8675, Japan

ABSTRACT The physiological properties of rice grains are immediately obvious to consumers. High-coverage metabolomic
characterization of the rice diversity research set predicted a negative correlation between fatty acid and lipid levels and
amylose/total starch ratio (amylose ratio), but the reason for this is unclear. To obtain new insight into the relationships
among the visual phenotypes of rice kernels, starch granule structures, amylose ratios, and metabolite changes, we inves-
tigated the metabolite changes of five Japonica cultivars with various amylose ratios and two knockout mutants (e1, a Starch
synthase IIIa (SSIIIa)-deficient mutant and the SSIIIa/starch branching enzyme (BE) double-knockout mutant 4019) by using
mass spectrometry-based metabolomics techniques. Scanning electron microscopy clearly showed that the two mutants had
unusual starch granule structures. The metabolomic compositions of two cultivars with high amylose ratios (Hoshiyutaka
and Yumetoiro) exhibited similar patterns, while that of the double-knockout mutant, which has an extremely high amylose
ratio, differed. Rice pedigree network analysis of the cultivars and the mutants provided insight into the association between
metabolic-trait properties and their underlying genetic basis in rice breeding in Japan. Multidimensional scaling analysis
revealed that the Hoshiyutaka and Yumetoiro cultivars were Indica-like, yet they are classified as Japonica subpopulations.
Exploring metabolomic traits is a powerful way to follow rice genetic traces and breeding history.

Key words: starch quality; traits; amylose ratio; metabolite profiling; pedigree networks; rice.

INTRODUCTION kernels (Choudhury and Juliano, 1980), and LPC content is

Rice is an important cereal of the human diet because it is
a good source of carbohydrates. Improvement in starch quality
in rice grains is directly connected to rice-grain eating and
cooking quality because 80–90% of the dry matter of rice
grains is composed of starch (Duan and Sun, 2005). Starch
granules in both rice and other cereals and crops consist of
amylose and amylopectin. Starch quality is affected by many
factors, including growth conditions and variety/cultivar dif-
ferences. Of these, the amylose/total starch ratio (amylose ra-
tio) is one of the most important factors for predicting the
physicochemical properties of starches (Blaszczak et al., 2003).
Rice grains contain lipids and free fatty acids (Prabhakar and
an important factor in the determination of starch quality
(Blaszczak et al., 2003; Hernández-Hernández et al., 2011).
Metabolomics can give a snapshot of biochemical status in
cells, body fluids, tissues, and organs in organisms (Fukushima
et al., 2009; Albinsky et al., 2010; Kusano et al., 2011). Metab-
olomics has provided great contributions to both phenotyping
and diagnostic analyses of plants (Nunes-Nesi et al., 2010). Gas
chromatography–mass spectrometry was applied for the
1
To whom correspondence should be addressed. E-mail mkusano005@psc.
riken.jp, tel. +81-45-503-9442, fax +81-45-503-9489.

Venkatesh, 1986; Proctor and Lam, 2001). These compounds
have an important effect on starch physicochemical properties
(Perez and Bertoft, 2010). Lysophosphatidylcholines (LPCs) are
the main phospholipids in cereal starches as well as in rice
ª The Author 2011. Published by the Molecular Plant Shanghai Editorial
Office in association with Oxford University Press on behalf of CSPB and
IPPE, SIBS, CAS.
doi: 10.1093/mp/ssr101, Advance Access publication 15 December 2011
Received 6 October 2011; accepted 13 November 2011

detection of primary metabolites such as fatty acids, sugars,
amino acids, and organic acids in metabolomics (Saito and Mat-
suda, 2010; Kusano et al., 2011). Lipidomic analysis including
phospholipids was performed using liquid chromatography–
mass spectrometry (LC–MS) (Welti et al., 2007; Murphy and Gas-
kell, 2011; Okazaki et al., 2011). An association analysis among
metabolite composition, genotype, and phenotype revealed
a close link between biomass and a specific combination of
metabolites including starch in recombinant inbred lines (Meyer
et al., 2007) and accessions in Arabidopsis (Sulpice et al., 2009).
Trait–metabolite association analysis using the rice diversity
research set (RDRS) predicted that amylose ratio and fatty acids
and other metabolites such as glycerol, phosphate, and putres-
cine are negatively correlated among rice cultivars, while the
two mutants, e1, a Starch synthase IIIa (SSIIIa)-deficient
mutant (amylose ratio, approximately 30%; Nipponbare
background (Fujita et al., 2007)), and the SSIIIa/starch branch-
ing enzyme (BE) double-knockout mutant 4019 (amylose
ratio, approximately 36%; Nipponbare/Kinmaze back-
ground), did not show the same trend (Redestig et al.,
2011). However, it is unclear why there was an association
between amylose ratio and the metabolites in rice kernels
of the cultivars but not in the mutants.
To obtain new insight about the relationships among visual
phenotypes including rice kernels, starch structure, amylose
ratio, and metabolite changes including phospholipids, we in-
vestigated the metabolite changes of Japonica cultivars with
various amylose ratios and two knockout mutants (Fujita
et al., 2007) by combing metabolite profile data obtained from
gas chromatography–time-of-flight–mass spectrometry (GC–
MS) and from LC–ion-trap–time-of-flight–MS (IT–MS) analyses.
We also conducted rice pedigree network analysis of the culti-
vars and the mutants to estimate the extent of the association
between metabolite-trait properties and their underlying
genetic basis in rice breeding in Japan.
RESULTS
Visual Phenotypes and Morphological Traits of the Rice
Seeds and Kernels
We observed the visual phenotypes of the rice seeds and
kernels of the five cultivars (Nipponbare, Kinmaze, Soft158,
Yumetoiro, and Hoshiyutaka) and the two knockout mutants
(e1 and 4019). Cultivar Yumetoiro and Hoshiyutaka are known
to have high amylose ratios, while Soft158 has a low amylose
ratio (Sakai et al., 1989; Ohta et al., 2004; Redestig et al., 2011).
Rice seeds and kernels of the two high-amylose cultivars showed
long-grain phenotypes (Figure 1D and 1E, and Table 1). The
rice seeds of other cultivars and the mutants looked similar
(Figure 1). However, the kernels of the e1 mutant had white
cores (Fujita et al., 2007) while the 4019 kernels looked almost
completely opaque (Figure 1F and 1G). The weight of 100 kernels
of the high-amylose cultivars and 4019 was low compared to that
of Nipponbare (Table 1).
Kusano et al. d Metabolite Profiling of Rice Kernels | 443
Figure 1. Visual Phenotypes of Rice Seeds and Brown Rice of the
Five Cultivars and the Two Knockout Mutants.
(A) Nipponbare, (B) Kinmaze, (C) Soft158, (D) Yumetoiro, (E) Hosh-
iyutaka, (F) e1, and (G) 4019.
Scale bar (white) = 5 mm.
Starch Granule Structures of the Cultivars and the
Knockout Mutants
To obtain insight about the relationships between amylose ra-
tio and starch granule structure, we conducted starch granule
imaging using scanning electron microscopy (SEM) (Figure 2).
SEM images of the starch granule structures of the cultivars
looked similar (Figure 2A–2E), while that of e1 showed relatively
small starch granules, some of which were round (Figure 2F and
2H). Kernels of the amylose-hyperaccumulating mutant
4019 had uniquely shaped starch granules, namely large and
spherical or wormlike (Figure 2G and 2I). The SEM imaging
results suggested that there is probably no correlation between
amylose ratio and starch granule structure, at least among the
assayed cultivars.
We further investigated how many starch granules are
packed in rice kernels of e1, 4019, and Nipponbare by using
SEM (Figure 3). The result of the cross-sections of endosperm
of e1 and 4019 clearly showed that the starch granules of the
mutants were loosely packed in rice kernels.
Metabolite Profiling of the Cultivars and the Knockout
Mutants by Using GC–MS and IT–MS
We next performed metabolite profiling to investigate the
extent of the metabolite changes in kernels of the cultivars
and the knockout mutants by using GC–MS and IT–MS (Sup-
plemental Data 1 and 2). Differentially changed metabolites
compared to the control Nipponbare were visualized using
a heatmap generated using hierarchical cluster analysis
(HCA) (Figure 4). The metabolite changes found in the
444 | Kusano et al. d Metabolite Profiling of Rice Kernels

Table 1. Variation of Rice Grain Morphological Traits.

Cultivar Nipponbare Kinmaze Soft158 Yumetoiro Hoshiyutaka e1 4019

Seed length (mm) 7.4 6 0.11 7.3 6 0.07 6.9* 6 0.12 8.3** 6 0.15 8.3*** 6 0.08 7.5 6 0.16 7.7 6 0.24
Seed width (mm) 3.4 6 0.21 3.4 6 0.02 3.3 6 0.04 2.9** 6 0.04 2.7** 6 0.02 3.4 6 0.04 3.4 6 0.05
Seed size (mm2)a 24.7 6 1.39 24.4 6 0.37 22.6* 6 0.65 23.8 6 0.77 22.7* 6 0.21 25.6 6 0.83 26.1 6 1.10
The weight of 2405.9 6 91.39 2201.4 6 111.45 2366.0 6 27.95 2315.2 6 77.97 2072.4*** 6 32.85 2529.5 6 64.54 2187.2* 6 73.43
100 seeds (mg)
Kernel length (mm) 5.2 6 0.06 5.0** 6 0.05 5.0* 6 0.08 5.8* 6 0.15 6.0*** 6 0.07 5.4* 6 0.04 5.1* 6 0.09
Kernel width (mm) 2.9 6 0.10 2.8 6 0.06 2.8 6 0.05 2.5*** 6 0.04 2.4*** 6 0.03 2.9 6 0.01 2.8 6 0.01
Kernel size (mm2)a 14.7 6 1.00 15.5 6 0.09 14.4 6 0.59 14.5 6 0.06 14.6 6 0.23 14.0 6 0.45 14.0 6 0.24
The weight of 2000.4 6 81.42 1850.3 6 97.89 1916.9 6 19.26 1831.9* 6 56.66 1694.9*** 6 32.77 2060.6 6 40.44 1725.6** 6 48.81
100 kernels (mg)

Values are presented as the mean 6 standard deviation (SD). Ten grains of each biological replicate were used (number of biological replicates, n = 3).
Differences between Nipponbare and each cultivar or mutant analyzed using Welch’s t-test were statistically significant (* p , 0.05; ** p , 0.005;
*** p , 0.0005).
a Size was tentatively calculated by multiplying length and width of each seed or kernel.

Figure 2. Scanning Electron Microscopy (SEM) of the Starch Granules.

SEM images are magnified 10003 for starch granules of (A) Nipponbare, (B) Kinmaze, (C) Soft158, (D) Yumetoiro, (E) Hoshiyutaka, (F) e1,
and (G) 4019. SEM images are magnified 40003 for (H) e1 and (I) 4019.
Scale bar (white) = 10 lm at a magnification of 10003 and 5 lm at a magnification of 40003.

metabolite profiles of the double-knockout mutant 4019
showed a unique pattern (class III in Figure 4). On the other
hand, the single knockout mutant e1 was grouped into class
I. Two high-amylose cultivars (Hoshiyutaka and Yumetoiro)
had similar patterns according to changes in their metabolite
profiles (class II in Figure 4). Subsequently, we assessed the
extent of metabolite-level changes of the cultivars and the
mutants at the chosen statistical threshold (5% false discovery
rate and log2-fold change . j1j). A Venn diagram was used to
find the significantly changed metabolites in common or those
 Kusano et al. d Metabolite Profiling of Rice Kernels | 445

Figure 3. Cross-Sections of Endosperm in Rice Kernels.

SEM images are magnified 30003 for (A) and (C) Nipponbare, (B) e1,
and (D) 4019. The mutant kernels of e1 and those of Nipponbare
were harvested on 2005, while 4019 and the corresponding Nippon-
bare were on 2008.
Scale bar (white) = 5 lm at a magnification of 40003.

that differed in Hoshiyutaka and Yumetoiro (Figure 5 and Table

2). Metabolites that were differentially changed in e1 and 4019
were compared to investigate how much the lack of SSIIIa
affects metabolic alternations in rice kernels (Figure 5 and Table
2). In Hoshiyutaka and Yumetoiro, levels of eight of the metab-
olites were commonly decreased (Figure 5 and Table 2). The
level of 18:0-LPC was increased in the amylose-rich cultivars
(Table 2). Between e1 and 4019, the levels of many metabolites
were commonly increased such as fructose-6-phosphate, fruc-
tose, glucose, 18:0-LPC, and intermediates in the TCA cycle
(Figure 5 and Table 2), while the phytol level was signifi-
cantly decreased (Table 2). Approximately 50% of the metab-
olites were commonly changed in e1 and 4019 (Figure 5).
We investigated changes in the metabolite levels of fatty
acids, phosphatidylcholines (PCs), and LPCs to validate our pre-
diction of amylose ratio and fatty acid and lipid levels in our pre- Figure 4. Heatmap with Cluster Dendrogram of the Differ-
vious study (Supplemental Tables 1 and 3). The linoleate level entially Changed Metabolites by Hierarchical Cluster Analysis (HCA).
was negatively correlated to amylose ratio among the tradi- Annotated or identified metabolites using gas chromatography–
tional cultivars (Redestig et al., 2011). However, there are no cor- mass spectrometry (GC–MS) (99 metabolites) and ion trap–mass
spectrometry (IT–MS) (10 metabolites) appear in the rows and
relation relationships among other fatty acids, PCs, LPCs, and
the samples appear in the columns. HCA was performed using
amylose ratios in the assayed cultivars. The levels of three fatty the matrix (log2-fold change; cultivars or mutants versus control
acids (linoleate, oleate, and palmitate), two LPCs (16:0-lysoPC NB). The complete list of differentially changed metabolites is
and 18:0-lysoPC), and six PCs were significantly increased in rice available in Supplemental Data 1 and 2. Abbreviations: HCA,
kernels of 4019 (Supplemental Table 3). hierarchical cluster analysis; NB, Nipponbare; Kin, Kinmaze; Hoshi,
Hoshiyutaka; Yume, Yumetoiro.
Rice Pedigree Network Analysis Reflected the Origin of the
Metabolite Profile Patterns of the Cultivars history for the five cultivars using a rice characteristic database
Metabolite profiling analysis clearly showed that the meta- (Ohta et al., 2004) and the Plant Genetic Resources Search
bolite profiles of e1, Kinmaze, and Soft158 have similar pat- System in the NIASGBdb to obtain insight into the origins of
terns, while those of Hoshiyutaka and Yumetoiro are similar metabolite profile alternations throughout rice relations in
(Figure 4). We investigated the network of rice breeding Japan (see ‘Methods’) (Takeya et al., 2011). A total of 171
446 | Kusano et al. d Metabolite Profiling of Rice Kernels

Table 2. Common Changed Metabolites between Hoshiyutaka

and Yumetoiro and Between e1 and 4019.

(A) Common changed metabolites between Hoshiyutaka and

Yumetoiro.
Metabolite log2-FC in log2-FC in
changes Metabolite name Hoshi/NB FDR Yume/NB FDR

Increased 18:0-lysoPC 1.0 0.00 0.7 0.00

Decreased Serine –1.5 0.00 –3.2 0.00
Alanine, beta- –1.7 0.00 –3.3 0.00
Aspartic acid –1.1 0.00 –1.1 0.00
Propane, 1,3-diamino- –1.4 0.00 –1.5 0.00
Glutamine –2.2 0.00 –4.7 0.00
Fructose –1.4 0.00 –1.6 0.00
M000000_A217004– –2.1 0.00 –2.9 0.00
101_MST_2174.6_
EITTMS_
Pyroglutamate –1.5 0.00 –3.1 0.00

(B) Common changed metabolites between e1 and 4019

Metabolite log2-FC in log2-FC in

changes Metabolite name e1/NB FDR 4019/NB FDR

Increased Homoserine 1.3 0.00 1.4 0.00

Aspartic acid 2.0 0.00 3.8 0.00
Arabinose 2.2 0.00 3.1 0.00
Shikimic acid 1.7 0.00 3.3 0.00
Fructose 2.6 0.00 2.5 0.00
N-Acetyl-D- 1.1 0.00 1.1 0.00
Figure 5. Venn Diagram of Significantly Changed Metabolites in glucosamine
Common or Those that Differ in (A) Hoshiyutaka and Yumetoiro M000000_A217004– 2.0 0.00 2.5 0.00
and (B) e1 and 4019. 101_MST_2174.6_
The Venn diagram in (A) shows the number of commonly decreased EITTMS_
metabolites in Hoshiyutaka and Yumetoiro compared to Nippon- Fructose-6-phosphate 4.8 0.00 9.9 0.00
bare (NB), while the Venn diagram in (B) represents the number M000000_A237002– 1.3 0.00 2.0 0.00
of commonly increased metabolites in e1 and 4019. 101_MST_2370.2_
Significant levels were set at false discovery rate (FDR) , 0.05 discovery EITTMS_
rate and log2-fold change (FC) . j1j. See also Table 2 for more detail. M000000_A250001– 2.0 0.00 1.6 0.00
101_MST_2495.5_
cultivars were involved in the breeding history of generating EITTMS_
the cultivars and the mutants (Figure 6). Hoshiyutaka and PR_MST_Polyol 3.2 0.00 4.2 0.00
Yumetoiro are progenies of common Indica cultivars such as (Hexitol)_2539.5
IR8, Peta, and Taichungnative1 (TN1). Glucose 2.0 0.00 2.6 0.00
Soft158 and Nipponbare were generated by the crossing Glycerol 2.7 0.00 4.1 0.00
of Japonica cultivars. Kinmaze is an old cultivar in Japan (Ohta Citrate 1.2 0.00 2.7 0.00
et al., 2004) that exists in the middle of the pedigree network Isocitrate 1.2 0.00 2.4 0.00
(Figure 6). 18:0-lysoPC 1.3 0.00 2.4 0.00
The double-knockout 4019 was generated by crossing e1 in Decreased Phytol –1.6 0.00 –3.6 0.00
the Nipponbare background with be, which is lacking a BE, in
the Kinmaze background (Figure 6), although metabotype of Significant levels were set at FDR , 0.05 discovery rate and log2-fold
the mutant 4019 was very specific (Figure 5). change (FC) . j1j. FC, fold change; FDR, false discovery rate; NB,
Nipponbare; Kin, Kinmaze; Hoshi, Hoshiyutaka; Yume, Yumetoiro.

Indica–Japonica Differentiation by Metabotypes of
Kinmaze, Soft158, Hoshiyutaka, and Yumetoiro, and those
of the Cultivars in RDRS
Rice pedigree network analysis suggested that the specific
metabolite changes found in Hoshiyutaka and Yumetoiro
may reflect an Indica-like metabotype. To estimate the rela-
tionships between metabotype changes and the differences
of the Indica or Japonica type in O. sativa, we conducted mul-
tidimensional scaling (MDS) using a metabolite profile dataset
consisting of the normalized metabolite profiles of Kinmaze,
 Kusano et al. d Metabolite Profiling of Rice Kernels | 447

Figure 6. Rice Pedigree Network Analysis Based on Information of Rice Relationships in Japan.
Each node represents a cultivar. The orange node shows the cultivars used for the study, while the blue node displays single (e1 and be)
or double (4019) knockout mutants. The green nodes are the common ancestors of Hoshiyutaka and Yumetoiro. Two Indica cultivars,
Mudgo and O. nivara, are shown as pink nodes. Edges represent pedigrees in rice breeding history. The green edge shows a parent–child
relation (pcr), while the pink double line shows a pair relationship. The blue line and purple broken edges show mutations and selections,
respectively. TN1, Taichungnative1.

Soft158, Hoshiyutaka, and Yumetoiro and those of the Indica

and Japonica cultivars (non-glutinous rice) that were randomly
chosen from the RDRS (see ‘Methods’). The coordinate plot
of the MDS analysis demonstrated that the metabotypes of
Hoshiyutaka and Yumetoiro were closer to those of the Indica
cultivars, although both were classified as Japonica cultivars
(Figure 7). Furthermore, the metabotype of Hoshiyutaka
was located in the center of the plot.

DISCUSSION
Fatty Acids and LPCs Probably Pack Starch Granules in Rice
Kernels
We expected that a negative correlation relationship between
fatty acid/lipid levels and amylose ratio suggested by our pre-
diction in our earlier study are probably required to maintain
normal starch granules in rice kernels (Redestig et al., 2011). To
verify this hypothesis, we observed the starch granule struc-
tures of the cultivars and the mutants by using SEM. As in Figure 7. The Coordinate Plot of Multidimensional Scaling (MDS)
Analysis Using Normalized Metabolite Profiles of Four Cultivars
our hypothesis, the starch granules of e1 and 4019, which show
and the Eight Representative Cultivars Chosen from the Rice Diver-
no significant correlation between fatty acid/lipid levels and sity Research Set (RDRS).
amylose ratio, had unique structures (Figure 2). In particular, Euclidean distance and log2 ratio values compared to the metab-
the shapes of the starch granules in 4019 were similar to those olite profiles of Nipponbare were used (see ‘Methods’). I, Indica; J,
in high-amylose maize (Perez and Bertoft, 2010). Observations Japonica.
448 | Kusano et al. d Metabolite Profiling of Rice Kernels
of the cross-sections of endosperm in these mutant kernels
revealed that starch granules were loosely packed in both
mutants (Figure 3).
Cereal starches contain free fatty acids and LPCs, and these
compounds are associated with the amylose fraction (Morrison
et al., 1984). Amylose and the longest linear branches of amylo-
pectin develop inclusion complexes with fatty acids, monogly-
cerides, and LPCs (Toro-Vazquez et al., 2003). Addition of
exogenous LPCs to maize starch paste can enhance its thermal
stability, resulting in more stable starch quality against heat be-
cause of starch–LPC inclusion complex formation (Toro-Vazquez
et al., 2003; Hernández-Hernández et al., 2011). The fold
changes in the levels of oleate, palmitate, 16:0-LPC, and 18:0-
LPC in e1 and 4019 were higher than those in the amylose-rich
cultivars (Supplemental Table 3), while the levels of linoleate
and palmitate showed a negative correlation with amylose
ratio across the representative non-glutinous cultivars except
for the mutants (Redestig et al., 2011). These results suggest
that SSIIIa and BE are not only essential for maintenance of
starch granule structure, but also affect the metabolite com-
position of rice kernels. We must emphasize that simple starch
analyses cannot provide us with such detailed insights about
an importance of starch biosynthesis-related genes for rice
kernels.
The appearance of rice kernels of the knockout mutants
e1 and 4019 showed white cores and opacity, respectively
(Figure 1F and 1G). Temperature is an important factor in
the determination of rice grain quality, particularly during
the grain-filling stage. When rice plants are grown at high
temperature during this stage, the rice kernels have a chalky
appearance and reduced weight (Tashiro and Wardlaw, 1991;
Yamakawa et al., 2007). The endosperm of the chalky rice
kernel ripened under high-temperature conditions were
loosely packed with elliptical-shaped starch granules contain-
ing air spaces (Yamakawa et al., 2007). Furthermore, the levels
of many genes and metabolites involved in starch biosynthesis
and carbohydrate metabolism changed in the developing
endosperm (Yamakawa and Hakata, 2010). Of these, SSIIIa is
mainly expressed in developing rice endosperm (Hirose and
Terao, 2004; Dian et al., 2005; Ohdan et al., 2005), and induction
of SSIIIa in rice depends on temperature (Yamakawa et al., 2007;
Yamakawa and Hakata, 2010). Kernels of the near-isogenic
line CSSL50-1, which was derived from the original donor
IR24 (Indica) in the largely Asominori background (Japonica),
showed a chalky appearance and loosely packed endosperm
granules with air spaces. Transcript profiling of the near isogenic
line revealed differential changes in the expression levels of
genes involving carbohydrate metabolism, signal transduction,
and redox homeostasis compared to those in the control Asomo-
nori(Liuet al., 2010).Asthe formationofgrainchalkinessis influ-
enced by multiple factors including starch synthesis, starch
granule structure, and arrangement triggered by external
stresses or down-regulation of genes involved in starch
biosynthesis, investigations into fatty acid and LPC levels in
developing rice grains of the mutants and other mutants with
chalky or opaque phenotype should be completed soon to
obtain detailed insight into the underlying mechanisms of
starch granule packing in rice kernels.
Metabolite Profiling Is a Powerful Tool to Distinguish
Cultivars Precisely
In this study, we used the Japonica cultivars and the mutants
with the Japonica background as their direct parental lines.
The Indica and Japonica cultivars have distinctive morpho-
logical and agronomic traits as well as differences at the
molecular level, such as DNA restriction fragment length
polymorphism (Ebana et al., 2005; Zhang et al., 2009), simple
sequence repeats, and chloroplast sequence (McCouch et al.,
2005). Traits including potassium chlorate resistance, drought
resistance, apiculus hair length, cold sensitivity, and phenol
reaction have been used often for Indica–Japonica differen-
tiation, although the spectra of the variation of these traits
overlap in the cultivars (Morishima and Oka, 1981). Seed
lengths of many Indica cultivars exceed those of Japonica
cultivars. However, the probability of misclassification using
this trait was approximately 40% (Morishima and Oka,
1981).
Umemoto and colleagues reported that Japonica-type am-
ylopectin tends to contain short-unit chains with a degree of
polymerization (DP) < 11 and long-unit chains with DP > 25,
while Indica-type amylopectin has a tendency to consist of in-
termediate-size chains with 12 < DP < 24 and long-unit
chains with DP > 25 (Umemoto et al., 1999). As the levels of
the long-unit chains are similar in the two amylopectins, the
specific characteristics of amylopectins depend on the pres-
ence of the short-unit or intermediate-sized chains in rice
grains (Umemoto et al., 1999; Nakamura et al., 2002a,
2002b). On the basis of the chain length in the amylopectin
clusters, starches of rice cultivars cultivated in Asia can be
classified into two types: L (for Indica) and S (for Japonica)
(Nakamura et al., 2002a). The metabolite changes of the culti-
vars showed similar patterns (Figure 4). Visible phenotypes of
Hoshiyutaka and Yumetoiro have longer seeds and kernels
than others (Figure 1 and Table 1). However, both cultivars
belonged to S-type rice according to their DP values. Soft158
(low-amylose rice) and Nipponbare (medium-amylose rice)
were also classified as S-type rice (Horibata et al., 2004). Use
of a metabotype-based classification strategy can give us in-
sight into the underlying status of seed composition in rice
(Figure 7). Hoshiyutaka and Yumetoiro are indeed hybrid
rice of the Indica and Japonica cultivars. According to the
rice pedigree networks, ancestors of Hoshiyutaka are almost
all Japonica cultivars except for Indica cultivars Mudgo and
IR8 (Figure 6). Investigation of the extent of metabolite changes
using a collection of backcross-recombinant inbred lines be-
tween Indica and Japonica cultivars will be performed in a future
study. In summary, our results suggest that phenotype–
metabotype associations between Hoshiyutaka and Yumetoiro
are well coordinated across rice breeding history in Japan.
Large-scale metabotyping of 171 cultivars in Figure 6 would

provide evidence to follow the traces of the rice pedigrees
as a future study.
METHODS
Chemicals
All chemicals except for the isotope-labeled chemicals used for
the GC–MS analysis (Kusano et al., 2007) were purchased
from Wako (Osaka, Japan; www.wako-chem.co.jp/egaiyo/) or
Sigma-Aldrich (Tokyo, Japan; www.sigmaaldrich/japan.html).
Plant Materials
The five rice cultivars (Nipponbare, Kinmaze, Soft158, Hoshiyu-
taka, and Yumetoiro) and two knockout mutants (e1 and 4019)
from RDRS were used for this study. Growth and harvesting
were performed as previously described (Redestig et al., 2011).
Observation of Starch Granules Using SEM
Starch granules were prepared from polished rice by using the
cold-alkali method (Yamamoto et al., 1973, 1981). Purified
starch granules were coated with gold using a fine coater
(JEOL JFC-1200) for 120 s. The morphology of the starch gran-
ules was examined by SEM (JEOL-5600, Tokyo, Japan). SEM was
performed in secondary electron mode at 15 kV. For observa-
tion of cross-sections of endosperm, dried rice seeds were cut
across the short axis with a razor blade. The surface was sputter
coated with gold and observed using SEM.
Metabolite Profiling
Metabolite profiling using GC–MS and IT–MS was performed in
accordance with the metabolomics metadata (Redestig et al.,
2011). The data were log2 transformed for further analysis.
Statistical Data Analysis
Statistical analyses were performed using R v2.12.1 (www.
r-project.org/) and Microsoft Office Excel 2007. Differences
in the morphological traits of rice seeds and kernels between
Nipponbare and each cultivar or mutant were determined us-
ing Welch’s t-test (p , 0.05). The fold changes of all cultivars
and mutants were calculated by dividing by the mean value of
Nipponbare. The differentially accumulated metabolites be-
tween a cultivar and Nipponbare and between a mutant
and Nipponbare were detected using the LIMMA package
(Smyth, 2004), which includes false discovery rate (FDR) correc-
tion for multiple testing (Benjamini and Hochberg, 1995). We
identified metabolites with significant changes in metabolite
levels (the log2-fold change . j1j) and the corresponding FDR-
corrected p-values that were ,0.05.
We used the log2-fold change matrix for HCA and MDS anal-
ysis. HCA was performed using Cluster 3.0 (de Hoon et al., 2004),
and the results of HCA were visualized using Java TreeView
v1.1.6 (http://jtreeview.sourceforge.net/). We applied Euclid-
ean distance as similarity matrices for the metabolites and culti-
vars or mutants and the average linkage for clustering. Using
Euclidean distance as implemented in the ‘cmdscale’ function
Kusano et al. d Metabolite Profiling of Rice Kernels | 449
of the R software, we performed MDS analysis, which tries to
demonstrate the underlying structure of empirically acquired
data. We also used the log2-fold change values for this analysis.
Rice Pedigree Network Analysis
Rice pedigree network analysis was performed using the fol-
lowing procedure. First, information about the rice pedigrees
among Nipponbare, Kinmaze, Soft158, Hoshiyutaka, and
Yumetoiro was collected from a rice characteristic database
(http://ineweb.narcc.affrc.go.jp/) (Ohta et al., 2004) and
NIASGBdb (www.gene.affrc.go.jp/databases-plant_search_en.
php) (Takeya et al., 2011). The origins of e1 and 4019 were in-
vestigated on the basis of the literature (Fujita et al., 2007;
Redestig et al., 2011). Second, we categorized the rice relation-
ships into four categories: (1) parent–child relation (pcr); (2)
pair relation (pair); (3) mutations induced by ethyl methanesul-
fonate mutation (EMS), by N-methyl-N-nitrosourea mutation
(MNU), by endogenous retrotransposon Tos17 insertion
(Tos17) (Agrawal et al., 2001), by c-ray irradiation (gamma),
and by natural mutation (mutation); and (4) natural selection
(selection). The pedigree matrix was imported to Cytoscape
2.8.1 (www.cytoscape.org/), and then the rice pedigree net-
work was visualized using a hierarchical layout algorithm.
There were a total of 171 nodes and 317 edges in the network.
SUPPLEMENTARY DATA
Supplementary Data are available at Molecular Plant Online.
FUNDING
This work was partially supported by a Grant-in-Aid for Scientific
Research (B) from the Japan Society for the Promotion of Science
(19380007 to N.F.) and the Program for the Promotion of Basic and
Applied Research for Innovations in Bio-oriented Industry to N.F.
ACKNOWLEDGMENTS
We thank Koji Takano at the RIKEN Plant Science Center for his
technical assistance with the IT–MS analysis and Hiroki Asai at Akita
Prefectural University for his help with the starch preparation. No
conflict of interest declared.
REFERENCES
Agrawal, G.K., Yamazaki, M., Kobayashi, M., Hirochika, R.,
Miyao, A., and Hirochika, H. (2001). Screening of the rice vivip-
arous mutants generated by endogenous retrotransposon Tos17
insertion: tagging of a zeaxanthin epoxidase gene and a novel
ostatc gene. Plant Physiol. 125, 1248–1257.
Albinsky, D., et al. (2010). Metabolomic screening applied to
rice FOX Arabidopsis lines leads to the identification of
a gene-changing nitrogen metabolism. Mol. Plant. 3, 125–142.
Benjamini, Y., and Hochberg, Y. (1995). Controlling the false discov-
ery rate: a practical and powerful approach to multiple testing.
Journal of the Royal Statistical Society. B57, 289–300.
450 | Kusano et al. d Metabolite Profiling of Rice Kernels
Blaszczak, W., Fornal, J., Amarowicz, R., and Pegg, R.B. (2003). Lip-
ids of wheat, corn and potato starch. Journal of Food Lipids. 10,
301–312.
Choudhury, N.H., and Juliano, B.O. (1980). Effect of amylose con-
tent on the lipids of mature rice grain. Phytochemistry. 19,
1385–1389.
de Hoon, M.J., Imoto, S., Nolan, J., and Miyano, S. (2004). Open
source clustering software. Bioinformatics. 20, 1453–1454.
Dian, W., Jiang, H., and Wu, P. (2005). Evolution and expression
analysis of starch synthase III and IV in rice. J. Exp. Bot. 56,
623–632.
Duan, M., and Sun, S.S. (2005). Profiling the expression of genes
controlling rice grain quality. Plant Mol. Biol. 59, 165–178.
Ebana, K., Kojima, Y., Fukuoka, S., Nagamine, T., and Kawase, M.
(2005). Development of an RFLP-based rice diversity research
set of germplasm. Breeding Science. 55, 431–440.
Fujita, N., et al. (2007). Characterization of SSIIIa-deficient
mutants of rice: the function of SSIIIa and pleiotropic effects
by SSIIIa deficiency in the rice endosperm. Plant Physiol. 144,
2009–2023.
Fukushima, A., Kusano, M., Redestig, H., Arita, M., and Saito, K.
(2009). Integrated omics approaches in plant systems biology.
Curr. Opin. Chem. Biol. 13, 532–538.
Hernández-Hernández, E., et al. (2011). Synchrotron X-ray scatter-
ing analysis of the interaction between corn starch and an exog-
enous lipid during hydrothermal treatment. Journal of Cereal
Science. 54, 69–75.
Hirose, T., and Terao, T. (2004). A comprehensive expression analysis
of the starch synthase gene family in rice (Oryza sativa L.). Planta.
220, 9–16.
Horibata, T., Nakamoto, M., Fuwa, H., and Inouchi, N. (2004). Struc-
ture and physicochemical characteristics of endosperm starches
of rice cultivars recently bred in Japan. J. Appl. Glycosc. 51,
303–313.
Kusano, M., et al. (2007). Unbiased characterization of genotype-
dependent metabolic regulations by metabolomic approach in
Arabidopsis thaliana. BMC Syst. Biol. 1, 53.
Kusano, M., Fukushima, A., Redestig, H., and Saito, K. (2011).
Metabolomic approaches toward understanding nitrogen
metabolism in plants. J. Exp. Bot. 62, 1439–1453.
Liu, X., et al. (2010). Transcriptome analysis of grain-filling caryop-
ses reveals involvement of multiple regulatory pathways in
chalky grain formation in rice. BMC Genomics. 11, 730.
McCouch, S., Garris, A.J., Tai, T.H., Coburn, J., and Kresovich, S.
(2005). Genetic structure and diversity in Oryza sativa L. Genetics.
169, 1631–1638.
Meyer, R.C., et al. (2007). The metabolic signature related to high
plant growth rate in Arabidopsis thaliana. Proc. Natl Acad. Sci.
U S A. 104, 4759–4764.
Morishima, H., and Oka, H.I. (1981). Phylogenetic differentiation
of cultivated rice. 22. Numerical evaluation of the Indica-Ja-
ponica differentiation. Japanese Journal of Breeding. 31,
402–413.
Morrison, W.R., Milligan, T.P., and Azudin, M.N. (1984). A relation-
ship between the amylose and lipid contents of starches from
diploid cereals. Journal of Cereal Science. 2, 257–271.
Murphy, R.C., and Gaskell, S.J. (2011). New applications of mass
spectrometry in lipid analysis. J. Biol. Chem. 286, 25427–25433.
Nakamura, Y., Sakurai, A., Inaba, Y., Kimura, K., Iwasawa, N., and
Nagamine, T. (2002a). The fine structure of amylopectin in endo-
sperm from Asian cultivated rice can be largely classified into
two classes. Starch-Starke. 54, 117–131.
Nakamura, Y., Umemoto, T., Yano, M., Satoh, H., and Shomura, A.
(2002b). Mapping of a gene responsible for the difference in am-
ylopectin structure between japonica-type and indica-type rice
varieties. Theoretical and Applied Genetics. 104, 1–8.
Nunes-Nesi, A., Fernie, A.R., and Stitt, M. (2010). Metabolic and
signaling aspects underpinning the regulation of plant carbon
nitrogen interactions. Mol. Plant. 3, 973–996.
Ohdan, T., et al. (2005). Expression profiling of genes involved in
starch synthesis in sink and source organs of rice. J. Exp. Bot.
56, 3229–3244.
Ohta, H., Ando, I., and Imbe, T. (2004). Constraction of a rice char-
acteristic database. Breed Res. 6, 253.
Okazaki, Y., Kamide, Y., Hirai, M.Y., and Saito, K. (2011). Plant lip-
idomics based on hydrophilic interaction chromatography cou-
pled to ion trap time-of-flight mass spectrometry. Metabolomics.
in press.
Perez, S., and Bertoft, E. (2010). The molecular structures of starch
components and their contribution to the architecture of starch
granules: a comprehensive review. Starch-Starke. 62, 389–420.
Prabhakar, J.V., and Venkatesh, K.V.L. (1986). A simple chemical
method for stabilization of rice bran. Journal of the American
Oil Chemists Society. 63, 644–646.
Proctor, A., and Lam, H.S. (2001). Rapid methods for milled rice sur-
face total lipid and free fatty acid determination. Cereal Chem-
istry. 78, 498–499.
Redestig, H., et al. (2011). Exploring molecular backgrounds of
quality traits in rice by predictive models based on high-coverage
metabolomics. BMC Syst. Biol. 5, 176.
Saito, K., and Matsuda, F. (2010). Metabolomics for functional
genomics, systems biology, and biotechnology. Annu. Rev. Plant
Biol. 61, 463–489.
Sakai, M., Shinoda, H., Hoshino, T., and Okamoto, M. (1989). A high
yielding Indica-Japonica-hybrid rice variety Hoshiyutaka. Jarq-
Japan Agricultural Research Quarterly. 23, 81–85.
Smyth, G.K. (2004). Linear models and empirical bayes methods
for assessing differential expression in microarray experiments.
Statistical Applications in Genetics and Molecular Biology. 3,
Article 3.
Sulpice, R., et al. (2009). Starch as a major integrator in the regulation
of plant growth. Proc. Natl Acad. Sci. U S A. 106, 10348–10353.
Takeya, M., et al. (2011). NIASGBdb: NIAS Genebank databases for
genetic resources and plant disease information. Nucleic Acids
Res. 39, D1108–D1113.
Tashiro, T., and Wardlaw, I.F. (1991). The effect of high-temperature
on kernel dimensions and the type and occurrence of kernel
damage in rice. Australian Journal of Agricultural Research.
42, 485–496.
Toro-Vazquez, J.F., Gomez-Aldapa, C.A., Aragon-Pina, A., Brito-de la
Fuente, E., Dibildox-Alvarado, E., and Charo-Alonso, M. (2003).
Interaction of granular maize starch with lysophosphatidylcholine

evaluated by calorimetry, mechanical and microscopy analysis.
Journal of Cereal Science. 38, 269–279.
Umemoto, T., Nakamura, Y., Satoh, H., and Terashima, K. (1999).
Differences in amylopectin structure between two rice varieties
in relation to the effects of temperature during grain-filling.
Starch-Starke. 51, 58–62.
Welti, R., et al. (2007). Plant lipidomics: discerning biological func-
tion by profiling plant complex lipids using mass spectrometry.
Frontiers in Bioscience. 12, 2494–2506.
Yamakawa, H., and Hakata, M. (2010). Atlas of rice grain filling-
related metabolism under high temperature: joint analysis
of metabolome and transcriptome demonstrated inhibition
of starch accumulation and induction of amino acid accumu-
lation. Plant Cell Physiol. 51, 795–809.
Kusano et al. d Metabolite Profiling of Rice Kernels | 451
Yamakawa, H., Hirose, T., Kuroda, M., and Yamaguchi, T. (2007).
Comprehensive expression profiling of rice grain filling-related
genes under high temperature using DNA microarray. Plant
Physiol. 144, 258–277.
Yamamoto, K., Sawada, S., and Onogaki, I. (1973). Properties of rice
starch prepared by alkali method with various condition. J. Jap.
Soc. Starch Sci. 20, 99–104.
Yamamoto, K., Sawada, S., and Onogaki, I. (1981). Effects of quality
and quantity of alkali solution on the properties of rice starch. J.
Jap. Soc. Starch Sci. 28, 241–244.
Zhang, Q.F., Ouyang, Y.D., Chen, J.J., and Ding, J.H. (2009). Advan-
ces in the understanding of inter-subspecific hybrid sterility and
wide-compatibility in rice. Chinese Science Bulletin. 54,
2332–2341.