Genetic networks controlling structural outcome of glucosinolate activation across development.

Abstract

Most phenotypic variation present in natural populations is under polygenic control, largely determined by genetic variation at quantitative trait loci (QTLs). These genetic loci frequently interact with the environment, development, and each other, yet the importance of these interactions on the underlying genetic architecture of quantitative traits is not well characterized. To better study how epistasis and development may influence quantitative traits, we studied genetic variation in Arabidopsis glucosinolate activation using the moderately sized Bayreuth x Shahdara recombinant inbred population, in terms of number of lines. We identified QTLs for glucosinolate activation at three different developmental stages. Numerous QTLs showed developmental dependency, as well as a large epistatic network, centered on the previously cloned large-effect glucosinolate activation QTL, ESP. Analysis of Heterogeneous Inbred Families validated seven loci and all of the QTL x DPG (days post-germination) interactions tested, but was complicated by the extensive epistasis. A comparison of transcript accumulation data within 211 of these RILs showed an extensive overlap of gene expression QTLs for structural specifiers and their homologs with the identified glucosinolate activation loci. Finally, we were able to show that two of the QTLs are the result of whole-genome duplications of a glucosinolate activation gene cluster. These data reveal complex age-dependent regulation of structural outcomes and suggest that transcriptional regulation is associated with a significant portion of the underlying ontogenic variation and epistatic interactions in glucosinolate activation.

Full text

Page 1

Genetic Networks Controlling Structural Outcome of
Glucosinolate Activation across Development
Adam M. Wentzell1,2¤a, Ian Boeye2, Zhiyong Zhang2¤b, Daniel J. Kliebenstein1,2*
1Genetics Graduate Group, University of California Davis, Davis, California, United States of America, 2Department of Plant Sciences, University of California Davis, Davis,
California, United States of America
Abstract
Most phenotypic variation present in natural populations is under polygenic control, largely determined by genetic
variation at quantitative trait loci (QTLs). These genetic loci frequently interact with the environment, development, and
each other, yet the importance of these interactions on the underlying genetic architecture of quantitative traits is not well
characterized. To better study how epistasis and development may influence quantitative traits, we studied genetic
variation in Arabidopsis glucosinolate activation using the moderately sized Bayreuth6Shahdara recombinant inbred
population, in terms of number of lines. We identified QTLs for glucosinolate activation at three different developmental
stages. Numerous QTLs showed developmental dependency, as well as a large epistatic network, centered on the previously
cloned large-effect glucosinolate activation QTL, ESP. Analysis of Heterogeneous Inbred Families validated seven loci and all
of the QTL6DPG (days post-germination) interactions tested, but was complicated by the extensive epistasis. A comparison
of transcript accumulation data within 211 of these RILs showed an extensive overlap of gene expression QTLs for structural
specifiers and their homologs with the identified glucosinolate activation loci. Finally, we were able to show that two of the
QTLs are the result of whole-genome duplications of a glucosinolate activation gene cluster. These data reveal complex age-
dependent regulation of structural outcomes and suggest that transcriptional regulation is associated with a significant
portion of the underlying ontogenic variation and epistatic interactions in glucosinolate activation.
Citation: Wentzell AM, Boeye I, Zhang Z, Kliebenstein DJ (2008) Genetic Networks Controlling Structural Outcome of Glucosinolate Activation across
Development. PLoS Genet 4(10): e1000234. doi:10.1371/journal.pgen.1000234
Editor: Gregory P. Copenhaver, The University of North Carolina at Chapel Hill, United States of America
Received August 8, 2008; Accepted September 22, 2008; Published October 24, 2008
Copyright: ? 2008 Wentzell et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: Funding for this project was obtained by National Science Foundation grants DBI 0642481 and MCB 0323759 to DJK and DEB 0608516 to AMW and
DJK. The funders had no role in any aspect of preparing this manuscript.
Competing Interests: The authors have declared that no competing interests exist.
* E-mail: kliebenstein@ucdavis.edu
¤a Current address: Seminis Vegetable Seeds, Woodland, California, United States of America
¤b Current address: Performance Plants, Inc., Kingston, Ontario, Canada
Introduction
Most phenotypic variation present in natural populations is
under polygenic control, largely determined by genetic variation at
multiple quantitative trait loci (QTLs), which has motivated
considerable efforts to elucidate the genetic basis of these polygenic
traits [1–3]. A complete understanding of quantitative traits
necessitates identification of the underlying genes and their
associated additive, dominance, and epistatic effects [2]. In
addition, the underlying genetic architecture of many quantitative
traits may vary across development and different environments. As
such, a comprehensive description of a quantitative traits genetic
architecture requires analysis in several developmental or
environmental contexts to assess stability of the genetic architec-
ture [2,4,5].
QTL mapping, which measure the association of genetic markers
with phenotypic variation, is one of the most common approaches
for identifying loci and epistatic interactions controlling polygenic
inheritance [1]. Improved statistical models, marker technology,
and genomic resources have facilitated QTL mapping experiments
for a wide array of quantitative traits, ranging from development
and morphology to metabolism and disease resistance [2,4,5];
However, QTL mapping experiments are often limited to a single
stage in development and one or few environments. As a
consequence, there is little information available to answer the
question of how the underlying genetic architecture varies across
developmental and environmental contexts.
Accurate characterization of a quantitative trait’s underlying
genetic architecture is often limited by practical considerations
that limit the number of progeny included in a mapping analysis.
Small populations are especially problematic in the presence of
epistasis between QTLs, as the pair wise comparisons required to
detect these interactions rapidly exhausts the available genotypic
variation, leading to an underestimation of numbers of loci and
interactions, resulting in an incomplete picture of the genetic
architecture [6,7]. One common type of epistasis occurs when a
trait is controlled by one or few large effect loci and numerous
modifier QTLs of smaller effect, a situation frequently observed in
plant disease resistance [8–16]. In such systems, the effects of any
modifiers are most detectable in those lines containing the
appropriate allele at the large effect locus; however this reduces
by half the population in which to detect these smaller effect loci,
significantly reducing statistical power [5–7]. Thus, resolution of
the underlying genetic basis of complex traits requires the analysis
of large populations across different environments or develop-
mental stages [17,18].
To investigate how development and epistasis can interact to
control the variation in an adaptive trait, we studied the outcome
PLoS Genetics | www.plosgenetics.org1October 2008 | Volume 4 | Issue 10 | e1000234

Page 2

of glucosinolate activation within Arabidopsis thaliana using a
moderately sized recombinant inbred population. Glucosinolates
are the inert storage form of a two part phytochemical defense
system found throughout the Brassicaceae, where biologically active
structures are catabolically produced by the enzyme myrosinase
[19–21](Figure 1) . The particular structural outcome, as defined
by the chemical structure of the end product, of glucosinolate
activation plays an important role in plant defense against insect
herbivory [22–25], as well as the nutritional and flavor
characteristics of brassicaceous crops [26]. Further, the structural
outcome shows significant intraspecific diversity such that natural
accessions activate a glucosinolate to either a nitrile or isothiocy-
anate depending upon their genotype. Thus an improved
understanding of the genetic basis of variation in structural
outcomes has important potential implications in evolution and
ecology as well as nutrition and agriculture.
Glucosinolate activation in Arabidopsis provides an excellent
model for studying how development and epistasis influence
quantitative traits, with a molecularly characterized biochemical
pathway comprising demonstrated epistatic interactions and
developmental variation. During glucosinolate activation, the
myrosinase enzyme catabolically generates the unstable intermedi-
ate. The final structural outcome of subsequent rearrangement of
this unstable intermediate is influenced by the presence or absence
of various structural specifier proteins (Figure 1). The Epithiospecifier
Protein (ESP) and an as yet unidentified simple nitrile structural
specifier (AtNSP), promote the formation of simple and epithioni-
triles at the expense of the default isothiocyanate rearrangement via
two biochemically related yet separate rearrangements (Figure 1)
[22,23,27–31]. The Epithiospecifier Modifier (ESM1) epistatically
modulates ESP mediated epithionitrile and simple nitrile rearrange-
ments (Figure 1)[22,23]. The observation of quantitative variation
influencing the developmental regulation of glucosinolate activation
enabled us to explorethe stability of QTLs and epistatic interactions
across development [30,32].
We used the Bayreuth (Bay-0)6Shahdara (Sha) recombinant
inbred lines (RILs) [33] to map QTLs controlling the structural
outcome of glucosinolate activation in Arabidopsis. These parental
accessions contain genetic variation for ESP and ESM1 and differ
in developmental regulation of structural outcomes [32]. We
measured structural outcomes at 30, 35 and 42 days post
germination (DPG), and compared the resulting maps to assess
the stability of the underlying genetic architecture across
development. These DPG were chosen because day 30 represents
the end of logarithmic growth in all of the lines, or Stage 1.10,
while day 42 is one week away from the earliest RIL flowering
(Stage 5.10) in our environmental conditions [32,34,35]. Thus, we
can focus on developmental changes in what is typically
considered a static rosette and is also the tissue and stages where
lepidopteran insects predominate on Arabidopsis in the wild [36].
This analysis identified eleven loci and twelve pair wise epistatic
interactions influencing structural outcomes, as well as eight
different QTL6DPG interactions. Heterogeneous Inbred Families
(HIFs) differing only for their genotypes at each QTL locus
validated seven loci and all of the QTL6DPG interactions tested.
Author Summary
A principal interest in biology is to understand how natural
genetic variation translates into phenotypic variation. A
key component of this connection is how the genetic
variation interacts with other sources of variation, such as
environment (G6E), development (G6D), or other genetic
loci (G6G or epistasis). To analyze the molecular under-
pinnings of these quantitative genetics interaction terms,
we investigated the genetic architecture of an adaptive
trait, glucosinolate activation, in Arabidopsis thaliana
during the development of what is considered a static
mature rosette. Variation in glucosinolate activation was
principally controlled by epistatic and G6D interactions.
Epistatic interactions identified both Mendelian epistasis,
where regulatory loci controlled enzymatic loci, and
quantitative interactions between regulatory loci. G6D
appeared to involve master regulatory loci as determined
by trans-eQTL hotspot analysis. Finally, two common
glucosinolate activation QTLs appear to have evolved via
gene loss and sub-functionalization following quadrupli-
cation of an ancestral genomic fragment, potentially by
two whole-genome duplications. Thus, genomic duplica-
tion events may facilitate the formation of quantitative
genetic variation. This study provides insights into the
molecular basis of the link between genetic and pheno-
typic variation in a potentially adaptive trait.
Figure 1. Glucosinolate Activation and the Subsequent Structural Rearrangement. Myrosinase enzymes initiate glucosinolate activation by
hydrolyzing the thioglucose bond, generating an unstable aglycone intermediate (bracketed). This intermediate can spontaneously rearrange into
the isothiocyanate structure. Epithionitrile structural specifiers such as ESP can promote the formation of epithionitrile structures from glucosinolates
with a terminal double bond, and simple nitriles from all other glucosinolates. As yet unidentified Arabidopsis simple nitrile structural specifiers
(AtNSPs) promote the formation of simple nitrile structures from all glucosinolates. In parenthesis are the Bayreuth and Shahdara accession showing
their predominant glucosinolate activation form.
doi:10.1371/journal.pgen.1000234.g001
Genetic Networks For Glucosinolate Activation
PLoS Genetics | www.plosgenetics.org2October 2008 | Volume 4 | Issue 10 | e1000234

Page 3

The availability of transcript accumulation data within 211 of
these RILs [37] enabled comparison of expression QTLs (eQTLs)
for structural specifier genes and their homologs, which demon-
strated collocation of eQTL clusters with eleven of the identified
loci. These data reveal complex age dependent regulation of
structural outcomes and suggest that transcriptional regulation is
associated with a significant portion of the underlying variation,
and may explain the epistatic interactions described here.
Results
Structural Outcomes of Glucosinolate Activation in Bay-0
and Sha
Bay-0 and Sha contain different glucosinolates due to variation at
the GSL-AOP and GSL-Elong loci, such that Bay-0 has 3-
hydroxypropyl glucosinolate as its main short chain aliphatic
glucosinolate and Sha has but-3-enyl glucosinolate [38]. These two
accessions also differ in the structures they produce following
activation of these glucosinolates. Bay-0 lacks functional ESP and
produces mixtures of simple nitriles and isothiocyanates, depending
on the age of the plant [32]. In contrast Sha possesses a functional
allele of ESP and produces mixtures of epithionitriles, simple nitriles
and isothiocyanates (Figure 1). In agreement with previously
published analysis, interplanted Sha parental controls had an
increasing epithionitrile proportion during development for both
the exogenous (Figure 2A and C) and endogenous glucosinolate
substrates (Table S4)[32]. In contrast to Sha, the Bay-0 parent
showed little variation in the structural outcome of exogenous allyl
glucosinolate activation between 30, 35and42DPG(Figure 2Aand
C). The activation products for the endogenous 3-hydroxypropyl
glucosinolate produced in Bay-0 could not be detected in this
experiment. Thus, there is developmental variation in glucosinolate
activation between the Bay-0 and Sha parental accession which
allows us to investigate how the genetics of glucosinolate activation
interact with plant development.
Distribution of Structural Outcomes in the RIL Population
We measured the structural outcome of glucosinolate activation
using exogenous allyl glucosinolate in the Bay-06Sha RILs and
compared the trait distribution to the interplanted parental
Figure 2. Ontogenic variation of Structural Outcome in Glucosinolate Activation. GC-FID was used to measure the structural outcome
from exogenous allyl glucosinolate activation in the RILs and interplanted Bay-0 (dashed line) and Sha (solid line) parental accessions at 30, 35, and 42
DPG. For the parental analysis, averages with standard errors are shown. For the RILs, The percent simple nitriles were binned into 1 percent intervals
from 0 to 20 percent, and percent epithionitriles were binned in 5 percent intervals from 0 to 100 percent and the number of RILs per bin are shown.
For the RILs, black bars show the distribution as measured at 30 DPG, grey is 35 DPG and white at 42 DPG. A) Simple nitriles in Bay-0 (dashed) and Sha
(solid) parental accessions at 30, 35, and 42 DPG. B) Distribution of simple nitrile formation in the RILs. C) Epithionitrile formation in parental
accessions. D) Distribution of epithionitrile formation in the RILs.
doi:10.1371/journal.pgen.1000234.g002
Genetic Networks For Glucosinolate Activation
PLoS Genetics | www.plosgenetics.org3October 2008 | Volume 4 | Issue 10 | e1000234

Page 4

controls. The use of exogenous allyl glucosinolate allowed us to
mask effects of variation in glucosinolate biosynthesis and
accumulation. Considerable transgressive segregation was ob-
served for simple nitrile where there were numerous RILs with
values higher and lower than the Bay-0 or Sha parents (Figure 2B).
Further, there was transgressive segregation for epithionitrile as
evidenced by the RILs with a higher value than the Sha parent
(Figure 2D). This suggests that alleles promoting the formation of
each structure exist in both parents. This is particularly surprising
in the case of epithionitrile proportions, as the Bay-0 parent can
not produce epithionitriles due to a lack of functional ESP and as
such might not be expected to contain genes enhancing
epithionitrile formation.
This population includes 212 lines producing no epithionitrile
structures following activation of allyl glucosinolate (Figure 2D). This
is consistent with the previously observed requirement of a functional
ESP allele for epithionitrile production [27,30,32]. For the sub-
population of RILs with a functional allele of ESP, the distribution
gradually shifted towards increased epithionitrile production from 30
DPG to 42DPG, in a manner consistent with the epithionitrile
increase observed in the Sha parental controls (Figure 2C). However,
individual RILs showed possible differences in the both the direction
ofandmagnitudeofchangeinepithionitrileproductionfrom30to42
DPG, suggesting that there is genetic variation in age dependent
control of epithionitrile proportion from 30 to 42 DPG (Table S4).
These possible differences will however require identification of the
QTL and HIF validation to ensure that this is not random variation
around the mean.
The distribution of nitrile formation within the RILs did not
show a similar ontogenic shift, but there were numerous lines that
had no simple nitrile formation at 42 DPG (Figure 2B). This low or
undetectable amount of simple nitriles is connected to epithioni-
trile proportions approaching complete utilization of the glucosi-
nolate substrate in these RILs at this DPG. This difference in
ontogenic regulation supports the model of simple nitrile and
epithionitrile production as being independent processes compet-
ing for the same substrate.
Heritability of Structural Outcomes
To compare the underlying genetics controlling the biochem-
ically distinct nitrile and epithionitrile glucosinolate activation
outcomes, we estimated heritability for each glucosinolate
activation structure from each measured glucosinolate (Table 1).
Because we were spatially constrained to only a single measure-
ment of each RIL at each DPG, this heritability estimate includes
both RIL and RIL6DPG effects and environmental variance is
not perfectly controlled. The endogenous glucosinolates are each
limited to roughly one quarter of the RILs due to independent
assortment of the GSL.AOP and GSL.Elong biosynthetic loci [38].
Further, the 4-methylsulfinylbutyl glucosinolate which derives
from transgressive segregation at the GSL.AOP and GSL.Elong lacks
a terminal alkene functional group, and can only form simple
nitrile and isothiocyanate structures (Figure 1)[39,40]. For all
detectable exogenous and endogenous glucosinolates, the herita-
bility of simple nitrile and epithionitrile proportions was
approximately 50%, with the sole exception of epithionitrile from
but-3-enyl glucosinolate at 69%.
Structural Proportion QTLs
To identify loci controlling the diversity of structural outcomes in
Bay-0 and Sha, we independently mapped QTLs for all traits at 30,
35, and 42 DPG. Analysis of epithionitrile proportions for
exogenous allyl glucosinolate revealed nine loci (Figure 3A, Table
S3), including seven novel QTLs and the previously identified ESP
and ESM1 loci [22,23]. Three loci (ESP, GSL.Activ.II.13, and ESM1)
were detected in the RILs at all three DPG, although the additive
effect of ESM1 switched direction from promoting nitrile formation
at 42 DPG relative to promotingisothiocyanate formationat 30 and
35 DPG. In spite of the fact that Bay-0 lacks functional ESP, four
QTLs showed a positive impact of the Bay-0 allele on epithionitrile
production, which is consistent with the observed transgressive
segregation in the RILs (Figures 2D and 3A).
We identified ten loci affecting the proportion of simple nitrile
structures produced from exogenous allyl glucosinolate, including
ESP, ESM1 and eight novel loci (Figure 3B, Table S3). The
average allelic substitution effect of these QTLs was 27% and the
median was 20%. ESP, ESM1 and six of the novel loci overlapped
with epithionitrile production QTLs with all but ESP showing the
same direction of allelic effect upon epithionitrile and nitrile
production (Figure 3). Six of the simple nitrile proportion QTLs
were significant at a single DPG, three were detected at two
consecutive DPG, and one locus (GSL.Activ.II.13) was detected in
all three QTL maps. Increased simple nitrile proportions were
fairly evenly distributed between the Bay-0 and Sha alleles
(Figure 3B). One locus (GSL.Activ.IV.16) exhibited significant
additive effects in opposite directions at 30 and 35 DPG
(Figure 3B). Isothiocyanates identified a combination of the nitrile
and epithionitrile QTLs (Figure S1). This suggests that the genetic
architecture underlying glucosinolate activation is much more
complex than the two locus model previously assumed [22,41].
Endogenous Glucosinolate Activation
We proceeded to compare QTLs identified using the exogenous
allyl glucosinolate to those identified with the endogenous but-3-
enyl and 4-methylsulfinylbutyl glucosinolate, the two glucosino-
lates with the highest level of accumulation in this population. The
Table 1. Heritability of Structural Outcomes of Glucosinolate Activation.
Trait
P ValueType III Sums of Squares
Heritability
GenoDPG ModelGenoDPG
% Simple Nitriles
,0.0010.39713083.4 6603.817.150.5
% Epithionitriles
,0.001
,0.001874209.6427015.0166069.748.8
Butenyl % Simple Nitriles
,0.001
,0.001 54039.528550.719106.552.8
Butenyl % Epithionitriles
,0.001
,0.001124753.5 82196.315517.365.9
4MSO % Simple Nitrile
,0.001
,0.001 348841.6178484.9 58429.951.2
Geno represents the genotype term in the model. DPG is the days-post-germination factor in the model.
doi:10.1371/journal.pgen.1000234.t001
Genetic Networks For Glucosinolate Activation
PLoS Genetics | www.plosgenetics.org4October 2008 | Volume 4 | Issue 10 | e1000234

Page 5

analysis of structural outcomes of endogenous glucosinolate
activation is complicated by independent assortment at the
GSL.Elong and GSL.AOP biosynthetic loci, limiting each measur-
able endogenous glucosinolate to one quarter of these RILs
[38,42,43]. QTL analysis of endogenous but-3-enyl glucosinolate
activation detected three loci and nine epistatic interactions
affecting simple nitrile formation, and only ESP and three epistatic
interactions for epithionitrile proportions. All QTLs were
consistent with those observed using the exogenous allyl
glucosinolate (Figure 3A and Table S3).
QTL analysis of 4-methylsulfinylbutyl glucosinolate activation
identified three loci affecting simple nitrile formation, which were
Figure 3. QTLs Controlling the Structural Outcome of Glucosinolate Activation. The five Arabidopsis chromosomes are depicted as lines in
a pentagonal layout with roman numerals placed at the 0 cM position for each chromosome. Arrows to the outside of each chromosome show the
positions of the identified QTLs. Inside each arrow, 1 indicates that the QTL was detected at 30 DPG, 2 for 35 DPG, 3 for 42 DPG and F for all DPG.
Arrows for loci where the Bay-0 allele has a positive effect point away from the chromosomes while arrows for QTLs with negative allele substitution
values point inward. The QTLs are named according to the nomenclature used in the text. Significant epistatic interactions are illustrated with arrows
inside the pentagon connecting the interacting loci and numbered to indicate the DPG at which the interaction was detected. A) QTLs and epistasis
affecting epithionitrile proportions. B) QTLs and epistasis affecting simple nitrile proportions.
doi:10.1371/journal.pgen.1000234.g003
Genetic Networks For Glucosinolate Activation
PLoS Genetics | www.plosgenetics.org5October 2008 | Volume 4 | Issue 10 | e1000234