Content uploaded by Rattan S Yadav
Author content
All content in this area was uploaded by Rattan S Yadav
Content may be subject to copyright.
Reproduced from Crop Science. Published by Crop Science Society of America. All copyrights reserved.
CROP SCIENCE, VOL. 47, MAY–JUNE 2007 969
RESEARCH
D
various periods during the growing season
is a common occurrence in pearl millet [Pennisetum glaucum
(L.) R. Br.] (van Oosterom et al., 1996c; Eldin, 1993), and thus is
one of the major factors in uencing crop yield and yield stability
(van Oosterom et al., 1996a, 1996b). Despite this, few millet-breed-
ing programs practice speci c selection for either general adapta-
tion to moisture-limited environments or for drought tolerance
per se, because of the practical di culties of doing both. Yield-
based empirical selection for adaptation under naturally occurring
stress environments is highly problematic (Blum 1988) because of
the unpredictability of naturally occurring stress, the quantitative
nature of adaptation, and the consequent predominance of geno-
type × environment interaction variance in selection results, as
well as the impracticality of sampling a representative range of
naturally occurring stress patterns. Yield-based empirical selection
for adaptation is possible using managed stress environments (e.g.,
Quantitative Trait Loci for Grain Yield
in Pearl Millet under Variable Post owering
Moisture Conditions
F. R. Bidinger, T. Nepolean,* C. T. Hash, R. S. Yadav, and C. J. Howarth
ABSTRACT
Pearl millet marker-assisted selection (MAS)
programs targeting adaptation to variable post-
owering moisture environments would bene t
from quantitative trait loci (QTLs) that improve
grain yield across the full range of post ow-
ering moisture conditions, rather than just in
drought-stressed environments. This research
was undertaken to identify such QTLs from an
extensive (12-environment) phenotyping data
set that included both stressed and unstressed
post owering environments. Genetic materials
were test crosses of 79 F2–derived F4 progenies
from a mapping population based on a widely
adapted maintainer line (ICMB 841) × a post ow-
ering drought-tolerant maintainer (863B). Three
QTLs (on linkage group [LG] 2, LG 3, and LG 4)
were identi ed as primary candidates for MAS
for improved grain yield across variable post-
owering moisture environments. The QTLs on
LG 2 and LG 3 (the most promising) explained a
useful proportion (13–25%) of phenotypic vari-
ance for grain yield across environments. They
also co-mapped with QTLs for harvest index
across environments, and with QTLs for both
grain number and individual grain mass under
severe terminal stress. Neither had a signi -
cant QTL × environment interaction, indicating
that their predicted effects should occur across
a broad range of available moisture environ-
ments. We have estimated the bene ts in grain
yield and accompanying changes in yield com-
ponents and partitioning indices that would
be expected as a result of incorporating these
QTLs into other genetic backgrounds by MAS.
F.R. Bidinger, T. Nepolean, and C.T. Hash, International Crops Research
Inst. for the Semi-Arid Tropics (ICRISAT), Patancheru P.O., Andhra
Pradesh 502 324, Ind ia; and R.S. Yadav and C.J. Howarth, Inst. of Grassland
and Environmental Research, Aber ystwyth, SY23 3EB, UK. Received 15
July 2006. *Corresponding author (t.nepolean@cgiar.org).
Abbreviations: BLUP, best linear unbiased predictor; GRMA, indi-
vidual grain mass; GRNO, grain number; GRYLD, grain yield; HI,
harvest index; LG, linkage group; LOD, logarithm of odds; MAS,
marker-assisted selection; PNHI, panicle har vest index; QTL, quan-
titative trait locus; RFLP, restriction fragment length polymorphism;
SSR, simple sequence repeat.
Published in Crop Sci. 47:969–980 (2007).
doi: 10.2135/cropsci2006.07.0465
© Crop Science Society of America
677 S. Segoe Rd., Madison, WI 53711 USA
All rights reserved. No part of this period ical m ay be reproduced or transmitted in any
form or by a ny mean s, electronic or mechanical, includ ing photocopyi ng, recording,
or any inform ation storage and retrieva l system, wit hout permission in writi ng from
the publisher. Per mis sion for pr inti ng and for repri ntin g the material contained herein
has been obtained by the publisher.
Published online May 31, 2007 Published online May 31, 2007
Reproduced from Crop Science. Published by Crop Science Society of America. All copyrights reserved.
970 WWW.CROPS.ORG CROP SCIENCE, VOL. 47, MAY–JUNE 2007
Bidinger et al., 1987a), but requires considerable resources
and is limited to a few types of stress.
Selection for drought tolerance per se is limited by
the lack of proven selection criteria. Screening techniques
have been proposed for various traits thought to be related
to drought tolerance in pearl millet (Yadav and Weltzien-
Rattunde, 1999), but few attempts to actively select for
tolerance have been reported. Research on genotype yield
di erences under terminal (unrelieved post owering)
stress has suggested possible criteria for identifying toler-
ance to th is t y pe of stress (Bidi nger et a l., 1987 b; Fussell et
al., 1991). A n extensive evaluation of one such cr iter ion—
maintenance of panicle harvest index under stress—for
improved tolerance of terminal stress indicated that yield
gains of 5% per initial cycle(s) under terminal stress are
possible with this criterion (Bidinger et al., 2000). The
feasibility of using more basic physiological parameters as
selection criteria is limited in pearl millet by the available
information on such traits as selection criteria.
Our own recent e orts to improve drought tolerance
in pearl millet have focused on the identi cation of quan-
titative trait loci (QTLs) for grain yield or closely related
traits under terminal stress conditions (Yadav et al., 2002,
2004). Several QTLs have been identi ed and the process
of evaluating their e ectiveness is underway (Bidinger et
al., 2005; Serraj et al., 2005). The focus on drought tol-
erance in this work is linked to the relative ease in pearl
millet of marker-assisted backcross (MA BC) transfer of
speci c QTLs to improve speci c aspects of widely used
hybrid parental lines (Bidinger and Hash, 2004). Using
MABC to enhance the drought tolerance of proven paren-
tal lines allows the breeder to concentrate on this trait,
with the knowledge that the recurrent parents are oth-
erwise fully acceptable to the seed industry (Witcombe
and Hash, 2000). This should be an e ective, short-term
approach, provided the drought tolerance QTLs have suf-
cient expression in the hybrids of the recipient parental
lines to justify the expense, with no negative e ects on
other desirable traits.
Apart from this speci c application, the general breed-
ing requirement remains not simply drought tolerance, but
improved adaptation (as measured by grain yield) to the full
range of expected moisture conditions during grain ll-
ing. Quantitative trait loci that enhance traits speci cally
linked to grain yield across the full range of grain- lling
moisture environments would be more useful to a general
breeding program, as they could be deliberately retained
in segregating generations by marker-assisted selection
(MAS), while phenotypic selection was practiced under
stress-free conditions for desired agronomic characteris-
tics. This approach would thus enhance broad adaptation
to the full range of grain- lling moisture environments,
simultaneously with selection for overall worth. As in the
case of MABC for drought tolerance, however, the feasi-
bility of this application of MAS depends on the identi -
cation of QTLs for yield or for strongly linked traits that
are e ective across the full range of expected moisture
conditions during grain lling.
The objectives of this study were (i) to identify QTLs
with favorable e ects on grain yield or on closely linked
traits that would be e ective across a broad range of grain-
lling moisture environments, and (ii) to estimate the
probable e ects of these when used as additional selection
criteria (using MAS) during the segregating generations
of a conventional millet breeding program.
MATERIALS AND METHODS
Plant Materials
This study was based on 79 skeleton-mapped F2–derived F4
progenies from the mapping population bred from the cross of
single inbred plants selected from each of two adapted main-
tainer lines, ICMB 841 × 863B, which was used in an earlier
study by Yadav et al. (2004). These were test crossed to the
drought-susceptible restorer line PPMI 301 for eld phenotyp-
ing. Line ICMB 841 (Singh et al., 1990) is of North Indian
origin and the parent of several commercial hybrids released for
this area. It is regarded as widely adapted and productive but
its hybrids do not ll grain well under terminal drought stress.
Line 863B was bred from the West African Iniadi landrace
material (Andrews and Anand Kumar, 1996), which has dem-
onstrated excellent tolerance to terminal drought stress under
managed drought-stress conditions at ICRISAT.
Genotyping
The genotyping of the F2 plants from the ICMB 841 × 863B
mapping population and the construction of the linkage map
was done and described by Yadav et al. (2004). In this study,
collinear markers were removed from the analysis before the
map was constructed using Mapmaker/Exp 3.0 (Lander et
al., 1987). The map obtained spans a total length of 551 cM
and comprises 79 loci including 50 restriction fragment length
polymorphism (RFLP) loci (Liu et al., 1994) and 29 simple
sequence repeat (SSR) markers (Allouis et al., 2001; Qi et al.,
2001, 2004) distributed across the seven linkage groups (LGs)
as named by Liu et al. (1994). Minor changes in the positions of
closely linked markers (~1.0 cM) were obtained but otherwise
all markers mapped to the same positions as the previously pub-
lished map of this cross.
Phenotyping
Test - c r o ssed F4 progenies were evaluated in similarly managed
replicated trials in the dry-season drought nursery at ICRISAT,
Patancher u, India, during the years 1998, 1999, 2000, and 2001.
Phenotyping environments each year included a fully irrigated,
stress-free environment and two (early- and late-onset) post ow-
ering stress environments. Irrigation in the early-onset treatment
was terminated approximately 1 wk before owering of the main
shoot, to initiate the stress about mid- owering to a ect both
seed number and seed lling. Irrigation in the late-onset treat-
ment was terminated 7 to 10 d later to initiate the stress in early
to mid-grain lling to a ect primarily seed lling.
Reproduced from Crop Science. Published by Crop Science Society of America. All copyrights reserved.
CROP SCIENCE, VOL. 47, MAY–JUNE 2007 WWW.CROPS.ORG 971
replicate l in environment i, Bijl is the e ect of block j in replicate
l of environment i, Gk is the e ect of progeny k, (GE)ki is the
interaction of progeny k with environment i, and εijkl represents
the residual for the ijklth plot. Each e ect in Eq. [1], except for
µ and Ei, was treated as a normally distributed random variable
with an expected value of zero and a constant variance. Since
our major interest here was to assess the consistency of detected
QTLs across the three moisture environments, the environment
term Ei in Eq. [1] was factorially partitioned as Ei = Euv = Yu +
Mv + (YM)uv, with Eq. [1] accordingly expanded to extract the
progeny × moisture-environment interaction e ects (GM)kv.
Here Yu is the e ect of year u, Mv is the e ect of moisture envi-
ronment v, and (YM)uv represents their interaction, with mois-
ture-environment e ects considered xed. Plot-level data from
ea ch of t he 12 i ndiv idu a l environment s we re a nalyz ed u sing the
following linear mixed e ects model
yjkl = µ + Rl + Bjl + Gk + εjkl [2]
where the replicate e ect Rl, due to its small sample size of three,
was treated as xed. The residual diagnostic plots from ReML
analysis of both models indicated that the model assumptions of
normality and constant variance were reasonably well satis ed
for all ve traits.
Multi-environment QTL mapping was done using the best
linear unbiased predictors (BLUPs) corresponding to the prog-
eny term Gk as obtained from Eq. [1–2], following the method
of composite interval mapping as outlined in Utz et al. (2000)
and implemented in PLA BQTL (Utz and Melchinger, 1996).
The presence of a putative QTL in any interval was tested using
a logarithm of odds (LOD) threshold of 2.5.
RESULTS AND DISCUSSION
Moisture Environments Effects
The e ect of moisture environment was highly signi -
cant for all measured variables, as expected in a managed
stress environment (Table 1). Grain yields declined from
an average of 378 g m−2 under favorable conditions during
grain lling to 254 g m−2 in the moderate, late-onset stress
All trials were planted in alpha (incomplete block) designs
with nine plots per block, replicated thrice. Individual plots
were two rows by 4.0 m long by 0.6 m apart; net (harvested)
plot area was two rows by 3.0 by 0.6 m. Trials were uniformly
managed to maximize growth and grain yield within the limits
of the moisture treatments, as described in Yadav et al. (2004).
Data were recorded on a har vested-area basis for oven-dry bio-
mass and grain yields (GRYLD) and converted to a square-
meter basis. Individual grain mass (GRMA) was determined
from the weight of triplicate samples of 100 oven dry grains,
and grain number per square meter (GRNO) was estimated
from grain yield and individual grain mass. Harvest index (HI)
was calculated as the ratio of grain to biomass yield, and panicle
harvest index (PNHI) from the ratio of grain to total panicle
weights. The former was considered an index of the ability to
convert biomass to grain across moisture environments and the
latter as a measure of the ability to set and ll grains across
moisture environments (Bidinger, 2002).
Results of the 1998 and 1999 evaluations done in the terminal
stress treatments only (four of the total of 12 phenotyping environ-
ments reported here) were used by Yadav et al. (2004) to assess the
e ects of year, stress intensity, and tester on identi cation of QTLs
for terminal drought tolerance. This study used data from the full
set of phenotyping environments—4 yr and three post owering
moisture environments—to identify and assess QTLs for grain
yield, its two major components (GRNO and GRMA), and the
two indices of partitioning e ciency (HI and PNHI), across the
ful l range of post owering moisture environments.
Data Analysis
Plot-level data on each trait were subjected to a linear mixed
model analysis using ReML in GENSTAT (GENSTAT 6
Committee, 2002). The data across the 12 environments were
analyzed based on the following linear mixed e ects model:
yijkl = µ + Ei + Ril + Bijl + Gk + (GE)ki + εijkl [1]
where yijkl is the observation on the ijklth plot corresponding to
progeny k in block j of replicate l in environment i, µ is the gen-
eral mean, Ei is the e ect of environment i, Ril is the e ect of
Table 1. Magnitude and signifi cance of different sources of variation from the mixed model ReML analysis. Moisture environ-
ment (treated as a fi xed effect) data are Wald statistics. The remaining data are estimates of variance components (with their
SEs shown in parentheses) for the corresponding sources of variation (treated as random effects).
Source of variation df Grain yield Grain number Individual grain mass Harvest index Panicle harvest index
g m−2 no. m−2 mg —————————— % ——————————
Year (Y) 3 915 (935) 22 576 872 (20 167 454) 1.358 (1.792) 0 (0.9) 4.16 (5.09)
Moisture environment (M) 2 70.4*** 24.6*** 54.9*** 65.0*** 39.7***
Y × M 6 611 (357) 5 506 429 (3 254 345) 2.241 (1.308) 2.3 (1.35) 5.5 (3.22)
Replication within Y × M 24 33 (25) 471 276 (350 570) 0.003 (0.018) 0.4 (0.27) 0.12 (0.13)
Blocks within replications 612 169 (30)*** 1 958 537 (424 729)*** 0.242 (0.064)*** 1.25 (0.23)*** 1.17 (0.25)***
Progeny (P) 78 92 (27)*** 2 080 487 (558 443)*** 1.312 (0.268)*** 3.42 (0.7)*** 2.8 (0.62)***
P × Y 234 30 (25) 1 466 813 (495 932)** 0.735 (0.133)*** 1.16 (0.25)*** 0.92 (0.3)**
P × M 156 20 (22) 165 631 (343 878) 0 (0.062) 0.76 (0.22)*** 0.9 (0.29)***
P × Y × M 468 34 (44) 322 985 (727 140) 0.334 (0.135)* 0 (0.32) 0.47 (0.432)
Residual 1259 29.3*** 29.5*** 29.4*** 29.5*** 29.7***
*Signifi cant at P < 0.05.
**Signifi cant at P < 0.01.
***Signifi cant at P < 0.0 01.
Reproduced from Crop Science. Published by Crop Science Society of America. All copyrights reserved.
972 WWW.CROPS.ORG CROP SCIENCE, VOL. 47, MAY–JUNE 2007
to 170 g m−2 in the severe, early-onset stress (Table 2).
Moisture-environment e ects on GRYLD were mirrored
by similar e ects on all yield-related variables, but the
absolute reductions varied with trait. Both GRNO and
GRMA were more severely reduced in the early-onset
stress than they were in the late-onset stress, but the reduc-
tions in GRMA were greater in both treatments (38 and
23%) than were the reductions in GRNO (29 and 14%,
Table 2), as stress was more severe in both treatments dur-
ing the determination of GRMA than during the deter-
mination of GRNO. Reductions in both HI and PNHI
were similar in the late-onset stress (10%), but greater in
HI (28%) than in PNHI (20%) in the early-onset stress
(Table 2). This re ects the greater e ect of the early-onset
stress on productive tiller number, as HI is reduced by the
failure of later tillers to produce grain, which are included
in total biomass but make no contribution to grain yield,
where PNHI is based only on productive tillers.
None of the e ects of the other components of envi-
ronment in the analysis were signi cant, apart from block
e ects (Table 1). The climate of the peninsular Indian dry
season is su ciently stable, and the management of the
drought nursery was su ciently repeatable, that neither
year nor year × moisture environment was a signi cant
source of variation for any of the measured variables. The
high degree of blocking in the experimental design was
able to account for a signi cant part of the local eld vari-
ability (Table 1), which, given the size of the experiment,
would have been expected.
Progeny and Progeny × Moisture
Environment Interactions
Progeny di erences were signi cant for all variables
reported (Table 1), and the ranges in the progeny BLUPs
across all three moisture environments varied widely
(Table 2). Thus, despite the relatively small number of
mapped progenies used, plus the fact that 50% of the
genome of each test-crossed progeny was similar, the
variation among test-crossed progenies for yield and yield
components across a range of grain- lling moisture envi-
ronments was substantial. Di erences among progenies
under terminal stress were expected because of the dif-
ferential tolerance of such stress between the parents of the
mapping population, but the range in progeny values was
generally as large in the favorable environment as it was in
the drought-stressed ones (Table 2).
Progeny × moisture environment interactions were,
somewhat surprisingly, signi cant only for HI and PNHI
(Table 1). Di erences among related, elite genetic materi-
als in HI and PNHI are often small in the absence of stress,
but become greater as stress a ects seed set, seed lling,
and assimilate supply to the grain di erentially in di er-
ent genotypes; so interactions with moisture environment
are expected in these two variables. The lack of signi -
cant progeny × moisture environment interactions for
GRYLD, GRNO, or GRMA is surprising, and suggests
that the primary di erences among progenies for these
variables were constitutive ones, which were little a ected
by moisture environment during grain lling. This con-
clusion was supported by a strong similarity in ranking of
individual progeny test crosses across the three grain- ll-
ing moisture environments: rank correlation coe cients
for GRYLD were ≥0.88 and for GRNO and GRMA ≥0.97
(data not presented). This is a positive nding in the con-
text of this research, i.e., if potential yield di erences are
more important in determining realized yield in the stress
environments than is stress tolerance, the chance of nding
e ective across-environment QTLs are greater. However,
QTLs for traits related to stress tolerance are still of inter-
est, especially when these are signi cantly correlated to
grain yield in severe stress environments (Table 3).
Progeny × year interactions were signi cant for GRNO
and GRMA, but not for GRYLD (Table 1), which suggests
that the year e ects on the two major yield components
were of a compensatory nature, i.e., an increase in one was
Table 2. Ranges and standard errors of difference (SED) of test-
crossed F4 progeny bes t linear unbiased p redictors for gr ain yield
and its key components in dif ferent moisture environments.
Best linear unbiased predictors
Tra i t Favorable
conditions
Moderate,
late-onset
stress
Severe,
early-onset
stress
Across
moisture
environments
Grain yield (g m−2)
Mean 378 254 170 267
Minimum 359 230 150 248
Maximum 401 272 191 286
SE D (p rog eny ) 11.3 11.1 9. 89 8. 3
Grain number ( no. × 103 m−2)
Mean 41.2 35.3 29.4 35.3
Min imum 37.2 31.3 25.7 31.4
Maximum 44.5 38.5 32.2 38.4
SED (progeny) 1.61 1.61 1.21 1.49
Grain mass (mg)
Mean 9.3 7.2 5.8 7.4
Minimum 8.2 6.1 4.7 6.3
Maximum 10.2 8.0 6.7 8.3
SED (progeny) 0.24 0.26 0.27 0.26
Harvest index (%)
Mean 44.8 40.2 32.4 39.1
Minimum 40.9 34.9 28.2 34.7
Maximum 49.0 45.2 37.8 44.1
SED (progeny) 0.98 1.45 1.51 0.83
Panicle harvest index (%)
Mean 75.4 68.2 60.5 68.0
Minimum 71.6 61.5 55.4 62.9
Ma x i mu m 78. 3 71.4 6 5.0 71.4
SED (progeny) 0.82 1.33 1.82 0.91
Reproduced from Crop Science. Published by Crop Science Society of America. All copyrights reserved.
CROP SCIENCE, VOL. 47, MAY–JUNE 2007 WWW.CROPS.ORG 973
o set by a decrease in the other, with the net result that
GRYLD itself was una ected. Progeny × year interactions
were also signi cant for HI and PNHI, but again not asso-
ciated with a parallel interaction for GRYLD. Thus in both
cases (moisture environment and year), the environmental
response of HI and PNHI did not appear to be strongly
linked to the response of GRYLD.
Relationships of Yield
and Component Variables
The questions of the relationship of grain yield with both
the basic yield components and the indicators of stress
response (plus the e ects of moisture environment on the
relationships) were examined by correlation of grain yield
and the component and indicator variables within and
across moisture environments. The two basic yield com-
ponents (GRNO and GRMA) were signi cantly, but only
very modestly, related to grain yield across environments,
with variation in both components explaining approxi-
mately 20% of the variation in GRYLD (Table 3). As
grain- lling stress increased (i.e., earlier stress onset), the
importance of GRNO to GRYLD declined slightly, and
the importance of GRMA increased slightly (Table 3), but
in neither case were the changes striking. In contrast, the
across-environment relationships of GRYLD with both
HI and PNHI were considerably stronger, with variation
in HI explaining 44% of the variation in GRYLD and
that in PNHI explaining 52% (Table 3). In both cases,
the strength of the relationship of GRYLD and the stress
response indicator variable increased with increasing stress
severity; correlation coe cients increased from 0.51 to
0.76 in the case of HI and from 0.57 to 0.79 in the case of
PN HI (Table 3). This sugges t s that g enotype di erences in
HI and especially PNHI are useful indicators of genotype
di erences in adaptation to stress, as measured by geno-
type grain yield. It is not clear therefore why progeny ×
moisture environment interactions in HI and PNHI were
not re ected in parallel progeny × moisture environment
interactions in GRYLD. On the strength of the behavior
of the correlations of GRYLD with both HI and PNHI,
however, we mapped QTLs for these variables, along with
GRYLD, GRNO, and GRMA, both within and across
grain- lling moisture environments.
Quantitative Trait Loci for Grain
Yield and Component Traits
Grain Yield
Quantitative trait loci for mean GRYLD across moisture
environments were mapped on LG 2, LG 3, and LG 4
(Table 4). The GRYLD QTLs on LG 2 also mapped in each
of the three individual moisture environments. Those on
LG 3 and LG 4 mapped in just the late-onset stress envi-
ronment (Table 4), which suggests that they are primar-
ily associated with yield di erences under terminal stress.
Neither had a signi cant QTL × environment (Q × E)
interaction, however, so their bene t under terminal stress
is not likely to be at the cost of GRYLD in the absence of
stress, even if the e ect is greater in stress environments.
The LG 2 QTL (linked to markers Xpsm458–Xpsmp2059,
Genomic Region 2, Fig. 1) was the most interesting, as it
had substantial LOD scores in all three moisture environ-
ments (6.3−6.9) as well as across environments (7.9) and
accounted for a signi cant proportion of the phenotypic
variance for GRYLD in both the stress (27−38%) and the
stress-free (28%) environments, as well as in the means
across environments (25%, Table 4). The favorable allele at
this locus was from 863B. The LG 3 and LG 4 QTLs had
lower LOD scores and accounted for much smaller frac-
tions of the phenotypic variance for GRYLD, both in the
late-onset stress environment and across environments,
than did the LG 2 QTL. The favorable alleles at both of
these loci were from ICMB 841.
Grain Number
One strong QTL for GRNO was mapped on LG 1
(Genomic Region 1) across moisture environments, with
the favorable allele from ICMB 841 (Table 4). The LOD
scores in the three moisture environments ranged from 7.7
to 9.3 (7.8 for the means across environments); this QTL
accounted for 34% of phenotypic variation across envi-
ronments and between 34 and 41% in individual moisture
environments. Despite being detected in all environments,
and thus appearing to be a broadly e ective QTL (i.e., little
a ected by moisture environment) the Q × E analysis did
indicate a signi cant (P ≤ 0.05) interaction with environ-
ment (Table 4). Presumably this was a consequence of dif-
ferences in the magnitude of e ects in di erent moisture
environments, as no crossover interaction was observed.
Noncrossover Q × E interaction is commonly due to either
inconsistency in detection of QTLs in di erent environ-
ments or di erential levels of expression of QTLs in dif-
ferent environments (Veldboom and Lee, 1996; Austin
and Lee, 1998; Li et al., 2003). As potential GRNO is
largely determined before actual owering, when the stress
Table 3. Pearson correlation coeffi cients of the test-crossed
F4 progeny grain yield and yield components for individual
grain-fi lling moisture environment and for mean values
across all moisture environments.
Pearson correlation coef fi cients
Grain
yield vs.
Favorable
conditions
Moderate,
late-onset
stress
Severe,
early-onset
stress
Across
moisture
environments
Grain number 0.492** 0.443** 0.398** 0.445**
Grain mass 0.378** 0.495** 0.587** 0.493**
Harvest index 0.514** 0.702** 0.761** 0.661**
Panicle harvest
index 0.574** 0.742** 0.792** 0.724**
**Signifi cant at P < 0.01.
Reproduced from Crop Science. Published by Crop Science Society of America. All copyrights reserved.
974 WWW.CROPS.ORG CROP SCIENCE, VOL. 47, MAY–JUNE 2007
environment di erences began, the signi cant across-envi-
ronment e ect of a GRNO QTL was expected.
This QTL co-mapped with a GRYLD QTL for both
favorable and moderate stress environments, and with a
GRMA QTL in all three individual moisture environ-
ments (Table 4). The favorable allele in the case of both the
GRYLD and GRMA co-mapped QTLs was contributed
by 863B, however, not ICMB 841 as was the case for the
GR NO QT L . Thus the ICM B 841 a l lele a t this QTL m ay
be linked to a smaller grain size, as GRNO and GRMA
are commonly negatively correlated in the crop (Bid-
inger et al., 2001). Because the importance of GRMA to
GRYLD increases as stress becomes more severe (Table 3),
it is the 863B allele at this QTL, linked to larger GRMA,
which is likely to be the more favorable for maintaining
GRYLD under severe stress. Thus if this QTL were con-
sidered for use in yield improvements across a range of
moisture environments, it would be the 863B allele, with
its favorable e ect on GRMA under stress, not the ICMB
841 allele with its favorable e ect on GRNO, that would
be the choice for selection.
Individual Grain Mass
There were two QTLs for GRMA detected across moisture
environments, one on LG 1 and one on LG 3, one of which
(LG 3) had signi cant Q × E interaction (Table 4). The
QTL on LG 1 (Genomic Region 1, Fig. 1) explained a very
signi cant proportion (40%) of the phenotypic variance of
the mean GRMA across environments, with signi cant
e ects in both the stress-free and mild-stress environments,
where it explained ≥40% of the phenotypic variance for
GRMA, and a very strong e ect in the severe-stress envi-
ronments, where it explained 57% of the phenotypic vari-
ance for this trait (Table 4). In all cases, the favorable allele
was contributed by the drought-tolerant parent 863B. The
LG 3 QTL accounted for a much smaller proportion of the
mean variance in GRMA across environments (16%, Table
4). Its e ect was similar in the favorable and severe-stress
conditions (LOD scores of 3.0−4.3, with additive e ects
of the ICMB 841 allele of 0.2 mg grain−1) but it was not
detected in the moderate-stress environments. There was
also a moderately strong GRMA QTL in LG 2 detected in
the severe-stress environments, but it was not detected in
Table 4. Quantitative trait loci (QTLs) identifi ed for grain yield and yield component traits for individual grain-fi lling moisture
environments and for mean across moisture environments (R2
adj is the fraction of the phenotypic variation in the trait explained
by the individual QTL; the additive effect is half the difference between the genotypic values of the two homozygotes at the
locus in question; a positive sign of additive effect indicates 863B allele favors the QTL; probability of Q × E is the probability
of the QTL × moisture environment interaction occurring by chance).
Stress-free
environment
Late-stress
environment
Early-stress
environment
Across moisture
environments
Tra i t LG †Marker interval LOD‡
(R2
adj)
Additive
effect
LOD
(R2
adj)
Additive
effect
LOD
(R2
adj)
Additive
effect
LOD
(R2
adj)
Additive
effect
P
(Q × E)
Grain yield, g m−2 1Xpsmp2069–Xpsm756 2.5 (11.3) 4.73 2.7 (20.4) 5.7
2Xpsm458–Xpsmp2050 6.3 (28.2) 7.78 6.3 (26.5) 6.4 6.9 (37.6) 8.1 7.9 (24.7) 6.0 NS§
3Xpsm108–Xpsmp2214 2.8 (11.6) −3.7 3.1 (12.6) −3.7 NS
4Xpsm1003d–Xpsm1007c 3.5 (17.3) −4.6 2.9 (18.7) −5.0 0.01
7Xpsmp2224–Xpsm717 2.8 (12.1) 4.1
Grain number,
no. × 103 m−2 1Xpsm761–Xpsm756 7.7 (33 . 8) −13 24 7. 8 ( 3 3.8 ) −13 0 6 9.3 (4 0 .8) −13 49 7.8 (3 3 .5) −128 3 0.01
4Xpsm1003d–Xpsm1007c 4.2 (17.0) −760
Grain mass, mg 1Xpsm761–Xpsm756 7.9 (41.4) 0.4 6.8 (33.7) 0.4 6.9 (57.2) 0.4 7.8 (40.2) 0.4 NS
2Xpsm322–Xpsmp2059 6.6 (34.6) 0.3
3Xpsm108–Xpsmp2214 3.0 (16.7) −0.2 4.3 (17.6) −0.2 2.9 (16.2) −0.2 0.01
6Xpsm588–Xpsm713 3. 0 ( 6 .1) 0.1
Harvest index, %
2Xpsmp2066–
Xpsmp2059 2.7 (13.6) 0.7 4.6 (20.5) 1.0 10.0 (37.6) 1.7 8.1 (25.7) 1.1 0.01
3Xpsm108–Xpsmp2214 6.1 (18.1) −0.8 5.7 (23.0) −1.0 5.6 (16.6) −0.8 7.6 (29.2) −1.0 NS
4Xpsm1003d–Xpsm1007c 2.8 (6.7) −0.5
5Xpsmp2064– Xpsm318 3.1 (13.5) 0.9 2.6 (15.9) 0.7 0.01
7Xpsmp2224–Xpsm717 5 .5 (34.5 ) 1. 5 4.8 (31.7) 1. 5 3.4 (27.9) 1. 3 4.9 ( 29 .9 ) 1.2 0 .0 5
Panicle harvest
index, % 1Xpsm761–Xpsm756 3.1 (13.4) 0.7 6.4 (29.3) 1.3 5.7 (44.0) 1.8 5.3 (40.5) 1.3 0.01
2Xpsmp2059–
Xpsmp2050 9.7 (42.8) 1.7 6.2 (51.4) 2.0 6.0 (50.8) 1.6 0.01
3Xpsm108–Xpsmp2214 3.5 (19.9) −0.8 3.8 (14.0) −0.7 5.8 (25.0) −1.0 6.0 (25.2) −0.8 NS
6Xpsm588–Xpsm713 4.9 (17.3) 1.0 5.1 (18.4) 0.8 NS
†LG, linkage group.
‡LOD, logarithm of odds.
§NS, not signifi cant at 0.05.
Reproduced from Crop Science. Published by Crop Science Society of America. All copyrights reserved.
CROP SCIENCE, VOL. 47, MAY–JUNE 2007 WWW.CROPS.ORG 975
the other environments and not detected across environ-
ments (Table 4).
All three of these GRMA QTLs co-mapped with
QTLs for GRYLD, but only the co-mapped GRYLD
QTLs on LG 2 and LG 3 were detected across environ-
ments (Table 4). These two QTLs each accounted for a
useful proportion of the phenotypic variance for mean
GRYLD across all environments (25% by the LG 2 QTL
and 13% by the LG 3 QTL). The strongest e ects of the
LG 2 GRMA QTL on GRYLD was in the early-onset
stress environment, where it accounted for 35 and 38% of
the phenotypic variance for GRMA and GRYLD, respec-
tively (Table 4), con rming the e ectiveness of the LG 2
QTL under severe stress. The strong LG 1 GRMA QTL
had a smaller e ect on grain yield, despite its stronger
e ect on GRMA (Table 4).
Harvest Index
Harvest index in this experiment was used as a measure
of genotype adaptation to various moisture environments
during grain lling, expressed as the ability to maximize
partitioning of total biomass to grain yield, despite vary-
ing levels of current photosynthesis for grain lling. Four
QTLs were detected for across-environment HI, one each
on four of the seven pearl millet LGs (Table 4). Percent-
age of the phenotypic variance for across-environment HI
accounted for by the individual QTLs ranged from 15 to
30%. Except for the QTL on LG 3, the favorable alleles for
HI QTLs were contributed by 863B. Three of the four HI
QTLs were subject to signi cant Q × E interaction, con-
sistent with the signi cant genotype × environment inter-
action found for the trait itself (Table 1). Of these three, the
rst (on LG 2) appeared to be primarily a drought-toler-
ance QTL, as the proportion of variance in HI it explai ned
(14–38%) increased as moisture stress increased (Table 4).
This QTL co-mapped with QTLs for both GRYLD and
GRMA that showed similar patterns of e ects across the
three moisture environments. The second HI QTL (on LG
5) was detected only in one moisture environment (early-
onset stress) and across environments, and did not co-map
with across-environment QTLs for GRYLD, GRNO,
or GRMA (Table 4), and thus appears to be of second-
ary interest. The third HI QTL (on LG 7) accounted
for a slightly greater proportion of the phenotypic vari-
ance for HI in the stress-free environments (35%) than in
the stress environments or the mean across environments
Figure 1. Pearl millet linkage groups 1, 2, and 3, based on the F2 mapping population derived from ICMB 841 × 863B, with quantitative
trait locus peaks (indicated by arrow) for grain yield and linked traits, sharing common genomic regions (proposed for marker-assisted
selection for broader adaptability), mapped at various moisture environments. Loci positions are given in Haldane cM to the left of the
linkage groups (GRYLD: grain yield; GRMA: grain mass; HI: harvest index; PNHI: panicle harvest index).
Reproduced from Crop Science. Published by Crop Science Society of America. All copyrights reserved.
976 WWW.CROPS.ORG CROP SCIENCE, VOL. 47, MAY–JUNE 2007
(approximately 30%). It also did not co-map with QTL for
across-environment GRYLD or its components, however,
and is thus also of secondary interest, despite its strong
e ect on HI. The one HI QTL that was not subject to Q ×
E interaction, on LG 3, accounted for a similar proportion
of the phenotypic variance (17–23%) in all three moisture
environments, and 29% of the variance in the mean across
all environments (Table 4). This HI QTL also co-mapped
with QTLs for GRMA and GRYLD, which also showed
a similarly consistent e ects in all the environments in
which the HI QTL was mapped.
The t wo most useful of the HI QTLs thus appear to be
those on LG 2 and LG 3, both of which appear to a ect grain
lling (GRMA), and both of which co-map with GRYLD
QTLs but have di erent environmental expressions. The
LG 2 HI QTL appears to be primarily a drought-toler-
ance QTL, with a signi cant Q × E interaction, but still
with a strong e ect on the mean across environments, and
the favorable allele from 863B. The LG 3 HI QTL has less
environment-speci c e ects (no Q × E interaction), with
moderate e ects in the individual moisture environments
and a strong e ect across environments, and with the favor-
able allele contributed by ICMB 841.
Panicle Harvest Index
Panicle harvest index in this experiment was used spe-
ci cally as a measure of tolerance to terminal drought,
expressed as the ability to set and ll grains under lim-
ited moisture. Four across-moisture-environment PNHI
QTLs were identi ed, of which two, on LG 1 and LG 2,
each accounted for nearly half of the phenotypic variance
in PNHI across environments (Table 4). Both of these
appeared to be primarily stress-tolerance QTLs, account-
ing for a higher proportion of phenotypic variance in the
stress environments than in the stress-free ones, and both
showing signi cant Q × E interactions. The favorable
allele at both QTLs was from 863B. The LG 1 PNHI
QTL co-mapped with an across-environment GRMA
QTL, which also expressed more strongly in the severe-
stress environments, but which did not show a signi cant
Q × E interaction (Table 4). The LG 2 PNHI QTL co-
mapped with across-environment QTLs for GRYLD and
HI, which similarly expressed more strongly in the stress
environments. The co-mapping of PNHI QTL with
both GRMA QTL and HI QTL is not unexpected under
grain- lling moisture stress.
The other two across-environment PNHI QTLs
(LG 3 and LG 6) did not have signi cant Q × E interac-
tions, but accounted for a lower proportion (25 and 18%
respectively) of the mean phenotypic variance in PNHI
across environments than did the QTLs on LG 1 and LG 2
(Table 4). The PNHI QTL on LG 3, for which the favor-
able allele was from ICMB 841, was detected in all three
moisture environments as well as across environments. It
co-mapped with across-environment QTLs for GRYLD,
HI, and GRMA, but only the HI QTL also had a nonsig-
ni cant Q × E interaction (Table 4). The PNHI QTL on
LG 6, with the favorable allele from 863B, was detected
only in the severe-stress environment and across environ-
ments. It co-mapped with a QTL for GRMA, which also
expressed only under severe-stress conditions.
Relationships to Previously
Identifi ed Quantitative Trait Loci
Using the same mapping population and genotyping data,
with only 2 yr of stress environment phenotyping data but
two testers, Yadav et al. (2004) mapped QTLs for grain
yield and yield-related traits under terminal stress (indi-
cated as stress-tolerance QTLs). The QTLs for GRYLD on
LG 1 and LG 2, for HI on LG 2 and LG 3, and for PNHI
on LG 1, LG 2, and LG 3 were identi ed in both studies,
with similar positions on these LGs, despite the di erent
QTL mapping programs used in this study (PLABQTL
1.1) and the study of Yadav et al. (2004) (Mapmaker/QTL
1.1). The two studies di ered in the nature of Q × E inter-
actions found, however. In the earlier study (Yadav et al.,
2004), QTLs were mapped for individual years of testing,
which identi ed some Q × E interactions that were of
the crossover type. In this study, all Q × E interactions
were due to di erences in the magnitude of QTL e ects
in di erent environments, despite the fact that this study
included a wider range of moisture environments than
did the earlier study. The QTLs with non-crossover Q ×
E interactions are preferable in a marker-based breeding
program, as the target allele has a similar e ect in di erent
environments (although these may di er in magnitude).
With crossover Q × E interactions, di erent alleles condi-
tion favorable performance in di erent environments, so
that a gain in one environment may be o set by a loss in a
contrasting environment.
This study also extends the information on the com-
mon QTLs identi ed in the two studies. The major LG
2 GRYLD QTL (863B allele) identi ed in both studies
is clearly not simply a drought-tolerance QTL, as earlier
reported, as it has a highly signi cant e ect on GRYLD
across all three moisture environments, including the
stress-free environments (Table 4). This QTL co-maps
with a strong QTL for HI across environments, suggest-
ing that its major e ect is a general increase in partition-
ing of biomass to grain. The general e ect of this QTL
on partitioning is also expressed under terminal stress in
terms of a highly signi cant e ect on PNHI (explaining
40–50% of the phenotypic variance), which underlines
the particular value of better partitioning to grain under
conditions of limited assimilate availability.
This study has also clari ed the nature of and the rea-
sons for the limited utility of the QTL on LG 1 (Table
4). The ICMB 841 allele at this QTL has a signi cant
Reproduced from Crop Science. Published by Crop Science Society of America. All copyrights reserved.
CROP SCIENCE, VOL. 47, MAY–JUNE 2007 WWW.CROPS.ORG 977
and consistent e ect on GRNO both in the individual
environments and across all environments (accounting for
34–41% of the variation for this trait). The increase in
grain number is linked to a smaller grain size, however,
which is expressed equally strongly in all three moisture
environments and across environments (accounting for
33–57% of the variation for this trait)—hence the failure
of this QTL to express as a GRYLD QTL across environ-
ments. In the severe-stress environment, the 863B allele at
this QTL accounted for a substantial portion (20%) of the
phenotypic variance for GRYLD through its very strong
e ect on GRMA. Across environments, however, this
QTL has limited value (although the 863B allele at least
had no negative e ects on GRYLD in any of the three
moisture environments).
The LG 3 HI and PNHI QTL detected in this study
was also detected in the study of Yadav et al. (2004), but
only across environments, and only for these two traits.
This study strengthens the value of this QTL consider-
ably, as it is now shown to have a signi cant e ect on both
traits in all three moisture environments, as well across
environments, plus a signi cant (if modest) e ect on both
GRMA and GRYLD across environments. Finally, this
study has identi ed two additional across-environment
QTLs (LG 4 and LG 6), one of which may be of value
in MAS. The LG 4 GRYLD QTL, with the favorable
allele f rom ICMB 841, account s for ver y usef u l 19% of the
variation in this trait across environments (Table 4). This
QTL does not co-map with any other observed trait across
environments, so it is not clear how it achieves its e ect on
GRYLD. Despite its signi cant Q × E interaction, how-
ever, the favorable allele from widely accepted ICMB 841
is of interest for MAS. The LG 6 PNHI and GRMA QTL,
with the favorable allele from 863B, expresses mainly in
the severe-stress environment and thus appears to be a sec-
ondary drought-tolerance QTL (Table 4).
Selection of Quantitative Trait
Loci for Marker-Assisted Selection
Marker-assisted selection is clearly limited to a small
number of target QTLs, as population sizes and costs
for marker analyses increase signi cantly as target QTL
numbers increase. Therefore it is important to select QTL
targets for MAS on the basis of as much information as
possible. The primary considerations in selection of tar-
get QTLs are (i) their likely direct (on the target trait)
and indirect (on other traits) e ects, and (ii) the expected
stability of expression of the QTL across environments.
The likely direct e ects of selection for the favorable allele
(in a homozygous condition) at a target QTL can be esti-
mated as twice the additive e ect of the allele, as deter-
mined from the analysis of the mapping population data.
The QTL e ect was estimated for both the main and co-
mapped traits a ected by each QTL, for the unstressed
environment, the combined drought-stressed environ-
ments, and for all environments (Table 5). The expected
stability across environments was assessed from the pres-
ence and nature of Q × E interactions in multi-environ-
ment QTL analyses (Table 4).
Primary Quantitative Trait Loci
Based on the estimated e ects on target traits and the
absence of Q × E interaction (Table 4), the 863B allele at
the QTL on LG 2 is the rst target for MAS for adaptation
Table 5. Direct and indirect effects of candidate quantitative trait loci (QTLs) for the improvement of grain yield across a range
of grain-fi lling moisture environments under moderate to severe grain-fi lling drought stress and in the absence of stress.
Expected effects are based on the target QTL being homozygous for the positive allele, i.e., the expected effect will be twice
the additive effect of the allele (GRYLD: grain yield; GRMA: grain mass; HI: harvest index; PNHI: panicle harvest index).
Expected effect of selecting for target allele
Linkage
group
Tar g ete d ge n om i c
region
Source
allele
Across moisture
environments Grain-fi lling drought stress Stress-free environments
Primary QTLs
2 Genomic region 2
(Xpsm322–Xpsmp2059)
863B Increase in GRYLD of 12 g m−2,
in HI of 2.2%, in PNHI of 3.2%
Increase in GRYLD of ≤16 g m−2, in
GRMA of ≤0.6 mg grain−1, in HI of
≤3.4%, in PNHI of ≤4.0%
Increase of 16 g m−2 in
GRYLD and 1.4% in HI
3Xpsm108–Xpsmp2214 ICMB 841 Increase in GRYLD of 7.4 g m−2,
in GRMA of 0.4 mg grain−1, in HI
of 2.0%, in PNHI of 1.6%
Increase in GRYLD of ≤7.4 g m−2, in
GRMA of ≤0.4 mg grain−1, in HI and
PNHI of ≤2%
Increase in GRMA of
0.4 mg grain−1, in HI and
PNHI of 1.6%
4Xpsm1003d–Xpsm1007c ICMB 841 Increase in GRYLD of 10 g m−2,
but no effect on other traits
Increase of ≤ 9 g m−2 in GRYLD and
≤1% in HI
No predicted effects
Secondary QTLs
1 Genomic region 1
(Xpsm761–Xpsm756)
863B Increase in PNHI of 2.6% and in
GRMA of 0.8 mg grain−1
Increase in PNHI of ≤3.6%, in GRMA of
0.8 mg grain−1, in GRYLD of ≤11 g m−2
Increase in PNHI of 1.4%, in
GRMA of 0.8 mg grain−1,
in GRYLD of 9 g m−2
7Xpsmp2224–Xpsm717 863B Increase in HI of 2.4% Increase in PNHI of 2.6–3.0%, in
GRYLD of ≤8 g m−2
Increase in HI of 3.0%
6Xpsm588–Xpsm713 863B Increase in PNHI of 1.6% Increase in PNHI of ≤2.0%, in GRMA
of ≤0.2 mg grain−1
No predicted effects
Reproduced from Crop Science. Published by Crop Science Society of America. All copyrights reserved.
978 WWW.CROPS.ORG CROP SCIENCE, VOL. 47, MAY–JUNE 2007
to varying moisture environments during grain lling.
This QTL was also identi ed by Yadav et al. (2004) for
improving terminal stress tolerance, based on its favor-
able e ects in terminal stress environments. The projected
e ect of selection for this QTL include an increase in
GRYLD of 12 g m−2 across all environments, plus major
gains in both GRMA and PNHI (Table 5). For individual
environments, incorporating this QTL results in a pre-
dicted gain of 16 g m
−2 in the absence of stress and 13
to 16 g m−2 in the terminal stress environments. This is
accompanied by predicted gains in HI (2.2%) and PNHI
(3.2%) across environments (Table 5).
The second target QTL for MAS would probably
be the ICMB 841 allele at either of the GRYLD QTLs
on LG 3 (linked to marker Xpsm108) or LG 4 (linked to
marker Xpsm1003d), both of which have a signi cant e ect
on across-environment GRYLD (Table 5). Despite their
being detected on the basis of across-environment e ects,
both appear to be primarily drought-tolerance QTLs, as
the only individual environment in which they had a sig-
ni cant e ect on GRYLD was the late-stress environment
(Table 4). Selection for the ICMB 841 allele at LG 3 has a
predicted e ect of increasing GRYLD by 7.4 g m−2 across
environments, as well as under moderate stress, and of
increasing HI and PNHI by between 1.6 and 2% (in abso-
lute terms) both across environments and in all individual
environments (Table 5). Selection for the ICMB 841 allele
at LG 4 had predicted e ects of increasing GRYLD by 10 g
m−2 across environments and by up to 9.2 g m−2 in the mod-
erate-stress environment, but was subject to a signi cant Q
× E interaction (Table 5). As both the LG 3 and LG 4 alleles
have generally similar predicted e ects on grain yield across
environments, the choice comes down to the improvement
in either individual environments, individual traits, or the
expected Q × E interaction. All of these factors favor the
LG 3 QTL. It has a predicted favorable e ect on both HI
and PNHI across environments, favorable e ects on several
traits in the stress-free environment, and nonsigni cant Q
× E interaction, none of which were true for the LG 4 QTL
(Tables 4 and 5).
Secondary Quantitative Trait Loci
There were several additional strong QTLs for across-
environment traits that did not have signi cant or direct
e ects on GRYLD in the test environments, despite the
positive correlations of grain yield and all of the traits for
which QTLs were identi ed (Table 3). The rst of such
QTLs is the GRMA and PNHI QTL on LG 1, with a
favorable allele from 863B. Marker-assisted selection for
this allele at this QTL predicted an increase in both traits
across environments, as well as in most individual envi-
ronments (Table 5); however, its e ect on GRYLD was
not consistent among individual environments. Its LOD
score (2.4) for GRYLD was just below the minimum nec-
essary to be considered as e ective across environments,
and its Q × E interaction would have been signi cant
had it been identi ed across environments. (The reason
is probably the o setting e ects of the alternate 863B and
ICMB 841 alleles at this locus, the former of which had
a bene cial e ect on GRMA and latter a bene cial e ect
on GRNO). Therefore this QTL would be less useful for
MAS than the primary QTLs above.
The next of the secondary QTLs is the strong across-
environ ment HI QTL on LG 7 (li nked to marker Xpsm717),
with the favorable allele from 863B. The predicted e ect
of selection for this QTL ranged from a gain of 2.6 to
3.0% in HI (i n absolute ter ms) in individua l environments
and 2.4% across moisture environments, and an increase
GRYLD by 8 g m−2 in the severe-stress environment,
where there was a strong correlation of HI and grain yield
(Table 3). This makes this QTL of potential interest for
strengthening adaptation to more serious stress environ-
ments (i.e., for improving drought tolerance). The third
of the secondary QTL is the relatively strong PNHI QTL
on LG 6 (linked to marker Xpsm588), with the favorable
allele again from 863B. Because of the large predicted gain
in PNHI from MAS for this QTL (Table 5), it may have
value for the improvement of partitioning to grain under
terminal stress, without a cost in stress-free environments;
however, this QTL would be of less interest for improving
GRYLD across a range of environments.
Two other criteria need to be considered in choosing
among potential target QTLs for MAS: (i) the availabil-
ity of easily scored marker polymorphism at loci anking
the QTL; and (ii) con dence in the QTL (related both
to the complexity of genetic and environmental factors
controlling the trait and to the power of the QTL detec-
tion experiment [mapping population size, marker num-
bers, and marker distribution]). Both of the primary QTLs
(LG 2 and LG 3) are acceptable on both criteria. Genomic
Region 2 (LG 2 for GRYLD, HI, and PNHI) has several
closely linked markers (Xpsm322–Xpsmp2050, spanning
the length of 14 cM) and a high level (as much as 0.92)
of polymorphic information content (PIC) for the mic-
rosatellite markers (Qi et al., 2004), making this region
very amenable to high-throughput genotyping. The LG 3
GRYLD, GRMA, HI, and PNHI QTL has linked marker
Xpsm18, which is an RFLP, but it maps to almost the same
position as SSR locus Xpsmp2070 (PIC of 0.90), so again
this QTL also should be very amenable to MAS.
The pearl millet map is being constantly updated with
newly developed SSR, expressed sequence tag SSR, sin-
gle-strand conformational polymorphism–single nucleo-
tide polymorphism, conserved-intron scanning primers,
and target region ampli ed polymorphism based markers
(C.T. Hash, personal communication, 2006). As a con-
sequence, highly polymorphic polymerase chain reaction
compatible markers will soon ank most of the RFLP
Reproduced from Crop Science. Published by Crop Science Society of America. All copyrights reserved.
CROP SCIENCE, VOL. 47, MAY–JUNE 2007 WWW.CROPS.ORG 979
markers that now anchor the pearl millet linkage map.
Use of these newly added markers will reduce costs and
time needed to exploit QTLs in targeted MAS programs
for variable grain- lling moisture conditions.
CONCLUSIONS
This study was intended to identify QTLs linked to
improved grain yield across a range of post owering
moisture environments, using data from a 4-yr phenotyp-
ing exercise done in a managed, eld drought nursery,
with annual stress-free, late-onset (moderate) terminal
stress, and early-onset (severe) terminal stress treatments.
The study was based on 79 skeleton mapped F2–derived
F4 progenies from the cross of two maintainer lines, one
widely adapted and one speci cally adapted to terminal
drought environments, crossed to a single tester. Data
were collected on grain yield, yield components, and crop
and panicle harvest indices. The phenotyping data set was
ideal for the purpose, as progeny and moisture-environ-
ment variances were highly signi cant but year, year ×
progeny, and year × environment variances were not.
Three QTLs (on LG 2, LG 3, and LG 4) were identi-
ed as primary candidates for MAS for improved grain yield
across variable post owering moisture environments. The
most promising of the three, those on LG 2 (863B allele) and
LG 3 (ICMB 841 allele) explained a useful proportion (25 and
13%, respectively) of phenotypic variance for GRYLD across
environments (Table 5). They also co-mapped with QTLs
for HI across environments, and with QTLs for GRMA,
PNHI, and HI under severe terminal stress. Neither had a
signi cant Q × E interaction, meaning that their predicted
e ects should occur across a broad range of available mois-
ture environments. Finally, both are linked to SSR mark-
ers so they are amenable to e cient MAS. The remaining
QTL (LG 4) is of secondary interest as it has a less consistent
performance across individual moisture environments and a
less clear e ect on secondary traits. Responses to MAS pre-
dicted for each of the identi ed GRYLD QTLs ranged from
7 to 10 g m−2 (70–100 kg ha−1) across environments and as
much 16 g m−2 in individual moisture environments. Finally,
this study has clari ed the action and utility of several of the
QTLs identi ed in an initial analysis of a subset (four of the
12 environments) of this data set (Yadav et al., 2004).
Acknowledgments
We would like to thank Mr. A. Ganapathi and Mr. P. Om Prakash for
assistance in producing the mapping population and testcross seed
of its progenies, and Mr. P.V.D. Maheshwar Rao for managing the
drought nursery in a very e cient manner and collecting the majority
of the eld data. This document is an output from projects (Plant
Sciences Research Programme R7375 and R8183) funded by the
UK Dep. for International Development (DFID) and administered
by CAZS Natural Resources for the bene t of developing countries.
The views expressed are not necessarily those of DFID.
References
Allouis, S., X. Qi, S. Lindup, M.D. Gale, and K.M. Devos. 2001.
Construction of BAC library of pearl millet, Pennisetum glau-
cum. Theor. Appl. Genet. 102:1200–1205.
Andrews, D.J., and K. Anand Kumar. 1996. Use of the West Afri-
ca n p earl m i llet landra ce In i ad i in cu ltiva r development. Pla nt
Genet. Resour. Newsl. 105:15–22.
Austin, D.F., and M. Lee. 1998. Detection of quantitative trait loci for
grain yield and yield components in maize across generations in
stress and nonstress environments. Crop Sci. 38:1296–1308.
Bidinger, F.R. 2002. Field screening for drought tolerance: Prin-
ciples and illustrations. p. 109−124. In N.P. Saxena and J.C.
O’Toole (ed.) Field screening for drought tolerance in crop
plants with emphasis on rice. ICRISAT, Patancheru, India.
Bidinger, F.R., S. Chandra, and V. Mahalakshmi. 2000. Genetic
improvement of tolerance to terminal drought stress in pearl
millet [Pennisetum glaucum (L.) R. Br.]. p. 59−63. In J.M. Rib-
aut and D. Poland (ed.) Molecular approaches for the genetic
improvement of cereals for stable production in water-limited
environments. Strategic Planning Workshop, CIMMYT, El
Batan, Mexico. 21−25 June 1999. CIMMYT, Mexico.
Bidinger, F.R., S. Chandra, and D.S. Raju. 2001. Genetic variation
in grain lling abilit y in dwart pearl millet [Pennisetum glaucum
(L.) R. Br.] restorer lines. Theor. Appl. Genet. 102:387–391.
Bidinger, F.R., and C.T. Hash. 2004. Pearl millet. p. 225−270. In
H.T. Nguyen and A. Blum (ed.) Physiology and biotechnology
integration for plant breeding. Marcel Dekker, New York.
Bidinger, F.R., V. Mahalakshimi, and G. Durga Prasada Rao.
1987a. A ss es sment of droug ht resist ance in pea rl mil let [Penni-
setum americanum (L.) Leeke]: I. Factors a ecting yields under
stress. Aust. J. Agric. Res. 38:37–48.
Bidinger, F.R., V. Mahalakshimi, and G. Durga Prasada Rao. 1987b.
Assessment of drought resistance in pearl millet [Pennisetum amer-
icanum (L.) Leeke]: II. Estimation of genotype response to stress.
Aust. J. Ag ric. Res. 38:49–59.
Bidinger, F.R., R. Serraj, S.M.H. Rizvi, C.J. Howarth, R.S. Yadav,
and C.T. Hash. 2005. Field evaluation of drought tolerance QTL
e ects of phenotype and adaptation in pearl millet [Pennisetym
glaucum (L.) R. Br.] topcross hybrids. Field Crops Res. 94:14–32.
Blum, A. 1988. Plant breeding for stress environments. CRC
Press, Boca Raton, FL.
Eldin, M. 1993. Analyse de l’e ect des de cits hydriques sur la
recolte du mi l au Niger, consequences agronom iques. p. 149–
160. In S. Hamon (ed.) Le mil en Afrique, diversity genet-
ique et agro-physiologique: Potentialites et contraintes pour
l’amelior iation et la culture. ORSTOM, Paris.
Fussell, L.K., F.R. Bidinger, and P. Bieler. 1991. Crop physiology
and breeding for drought tolerance: Research and develop-
ment. Field Crops Res. 27:183–199.
GENSTAT 6 Committee. 2002. GENSTAT 6 reference manual.
Clarendon Press, Oxford, UK.
Lander, E., P. Green, J. Abrahamson, A. Barlow, M. Daly, S. Lincoln,
and L. Newburg. 1987. MAPMAKER: An interactive computer
package for constr ucting primar y genetic linkage maps of exper-
imenta l and natural populations. Genomics 1:174–181.
Li, Z.K., S.B. Yu, H.R. La tte, N. Huang, B. Courtois, S. Hit-
talmani, C.H.M. Vijayakumar, G.F. Liu, G.C. Wang, H.E.
Shashidhar, J.Y. Zhuang, K.L. Zheng, V.P. Singh, J.S.
Sidhu, S. Sr ivantaneeyakul, and G.S. Khush. 2003. QTL ×
environment interactions in rice: I. Heading date and plant
height. Theor. Appl. Genet. 108:141–153.
Reproduced from Crop Science. Published by Crop Science Society of America. All copyrights reserved.
980 WWW.CROPS.ORG CROP SCIENCE, VOL. 47, MAY–JUNE 2007
Liu, C.J., J.R. Witcombe, T.S. Pittaway, M. Nash, C.T. Hash, C.S.
Busso, and M.D. Gale. 1994. An RFLP-based genetic linkage
map of pearl millet (Pennisetum glaucum). Theor. Appl. Genet.
89:481–487.
Qi, X., S. Lindup, T.S. Pittaway, S. Allouis, M.D. Gale, and K.M.
Devos. 2001. Development of simple sequence repeat mark-
ers from bacterial arti cial chromosomes without subcloning.
Biotechniques 31:355–362.
Qi, X., T.S. Pittaway, S. Lindup, H. Liu, E. Waterman, F.K. Padi,
C.T. Hash, J. Zhu, M.D. Gale, and K.M. Devos. 2004. An
integrated genetic map and a new set of simple sequence
repeat markers for pearl millet, Pennisetum glaucum. Theor.
Appl. Genet. 109:1485–1493.
Serraj, R., C.T. Hash, S.M.H. Rivzi, A. Sharma, R.S. Yaday, and F.R.
Bidinger. 2005. Recent advances in marker-assisted selection for
drought tolerance in pearl millet. Plant Prod. Sci. 8:334 –337.
Si ng h, S. D., P. S in gh , K .N. Ra i, and D .J. An dr ew s. 1990. Reg istratio n
of ICMA 841 and ICMB 841 pearl millet parental lines with A1
cytoplasmic-genetic male-sterility system. Crop Sci. 30:1378.
Utz, H.F., and A.E. Melchinger. 1996. PLABQTL: A prog ram
for composite interval mapping of QTL. Available at http://
wheat.pw.usda.gov/ja g/papers96/paper196/inde xp196.ht m l
(veri ed 15 Mar. 2007). J. Agric. Genomics 2:1.
Utz, H.F., A.E. Melchinger, and C.C. Schön. 2000. Bias and
sampling error of the estimated proportion of genotypic
variance explained by quantitative trait loci determined
from experimental data in maize using cross va lida-
tion and validation with independent samples. Genetics
154:1839−1849.
van Oosterom, E.J., F.R. Bidinger, V. Mahalakshmi, and K.P. Rao.
1996a . E ect of water availability pattern on yield of pearl m illet
in semi-arid tropical environments. Euphytica 89:163–174.
van Oosterom, E.J., V. Mahalakshmi, F.R. Bidinger, and K.P.
Rao. 1996b. E ect of water availability and temperature on
the genotype × environment interaction of pearl millet in
semi-arid tropical environments. Euphytica 89:175–184.
van Oosterom, E.J., M.L. Whitaker, and E. Weltzien-Rattunde.
1996c. Integrating genot ype × environment ana lysis, char-
acterisation of drought patter ns, and farmer preferences to
identify adaptive plant traits for pearl millet. p. 383 –402. In
M. Cooper and G.L. Hammer (ed.) Plant adaptation and crop
improvement. CAB Int., Wallingford, UK.
Veldboom, L.R., and M. Lee. 1996. Genetic mapping of quantita-
tive trait loci in maize in stress and nonstress environments: I.
Grain yield and yield components. Crop Sci. 36:1310–1319.
Witcombe, J.R., and C.T. Hash. 2000. Resistance gene deploy-
ment strategies in cereal hybrids using marker-assisted selec-
tion: Gene pyram iding, three-way hybrids, and synthetic
parent populations. Euphytica 112:175–186.
Yadav, O.P., and E. Weltzien-Rattunde. 1999. Breeding for adap-
tation to abiotic stress. p. 317–336. In I.S. Khairwal et al. (ed.)
Pearl millet breeding. Oxford and IBH, New Delhi, India.
Yadav, R.S., C.T. Hash, F.R. Bidinger, G.P. Cavan, and C.J.
Howarth. 2002. Quantitative trait loci associated with
traits deter mining grain and stover yield in pearl millet
under terminal drought stress conditions. Theor. Appl.
Genet. 104:67–83.
Yadav, R.S., C.T. Hash, F.R. Bidinger, K.M. Devos, and C.J.
Howarth. 2004. Genomic regions associated with grain
yield and aspects post owering drought tolerance in pearl
millet across stress environments and tester background.
Euphytica 136:265–277.
