More about quantitative trait locus mapping with diallel designs.
ABSTRACT We present a general regression-based method for mapping quantitative trait loci (QTL) by combining different populations derived from diallel designs. The model expresses, at any map position, the phenotypic value of each individual as a function of the specific-mean of the population to which the individual belongs, the additive and dominance effects of the alleles carried by the parents of that population and the probabilities of QTL genotypes conditional on those of neighbouring markers. Standard linear model procedures (ordinary or iteratively reweighted least-squares) are used for estimation and test of the parameters.
- [Show abstract] [Hide abstract]
ABSTRACT: The research area of association mapping (AM) is currently receiving major attention for genetic studies of quantitative traits in all major crops. However, the level of success and utility of AM achieved for crop improvement is not comparable to that in the area of human health care for diagnosis of complex human diseases. These AM studies in plants, as in humans, became possible due to the availability of DNA-based molecular markers and a variety of sophisticated statistical tools that are evolving on a regular basis. In this chapter, we first briefly review the significance of a variety of populations that are used in AM studies, then briefly describe the molecular markers and high-throughput genotyping strategies, and finally describe the approaches used for AM studies. The major part of the chapter is, however, devoted to analysis of reasons why the results of AM have been underutilized in plant breeding. We also examine the opportunities available and challenges faced while using AM for crop improvement programs. This includes a detailed discussion of the issues that have plagued AM studies, and the solutions that have become available to deal with these issues, so that in future, the results of AM studies may prove increasingly fruitful for crop improvement programs.Advances in genetics. 01/2014; 85C:109-147.
- [Show abstract] [Hide abstract]
ABSTRACT: QTL mapping in multiple families identifies trait-specific and pleiotropic QTL for biomass yield and plant height in triticale. Triticale shows a broad genetic variation for biomass yield which is of interest for a range of purposes, including bioenergy. Plant height is a major contributor to biomass yield and in this study, we investigated the genetic architecture underlying biomass yield and plant height by multiple-line cross QTL mapping. We employed 647 doubled haploid lines from four mapping populations that have been evaluated in four environments and genotyped with 1710 DArT markers. Twelve QTL were identified for plant height and nine for biomass yield which cross-validated explained 59.6 and 38.2 % of the genotypic variance, respectively. A major QTL for both traits was identified on chromosome 5R which likely corresponds to the dominant dwarfing gene Ddw1. In addition, we detected epistatic QTL for plant height and biomass yield which, however, contributed only little to the genetic architecture of the traits. In conclusion, our results demonstrate the potential of genomic approaches for a knowledge-based improvement of biomass yield in triticale.Theoretical and Applied Genetics 10/2013; 127(1). · 3.51 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: El objetivo de este trabajo fue evaluar marcadores moleculares y caracteres cuantitativos en un cruzamiento dialélico completo sin recíprocos, entre cinco líneas recombinantes de tomate y sus híbridos. Seobtuvieron perfiles de AFLP ("amplified fragment length polymorphism") y de polipéptidos del pericarpio en cuatro estados de madurez del fruto de 15genotipos. Seevaluaron, entre otros: peso, acidez titulable, pH, vida poscosecha y firmeza. Secalculó el porcentaje de polimorfismo para los marcadores moleculares y el porcentaje de variabilidad genética para los caracteres cuantitativos en el grupo de líneas recombinantes, el de híbridos y el conjunto de genotipos. Serealizaron análisis de agrupamiento con cada nivel de variación genética. Para AFLP, el porcentaje de polimorfismo varió entre 34 y 54% y, para los perfiles polipeptídicos, entre 40 y 78%. Mayor polimorfismo fue observado en el grupo de híbridos. La variabilidad genética fue de 100% para acidez y 34% para firmeza, con los mayores valores en los parentales. La similitud genética varió entre los genotipos según el nivel de variación genética; pero la consistencia en el agrupamiento de algunas líneas recombinantes y sus híbridos fue conservada, lo que evidenció asociaciones entre los datos moleculares y fenotípicos.Pesquisa Agropecuária Brasileira 05/2011; 46(5):508-515. · 0.68 Impact Factor
Genet. Res., Camb. (2000), 75, pp. 243–247. Printed in the United Kingdom ? 2000 Cambridge University Press
More about quantitative trait locus mapping with diallel
AHMED REBAI?* BRUNO GOFFINET
INRA Centre de Toulouse, Unit of Biometry and Artificial Intelligence, BP 27, 31326 Castanet-Tolosan, France
(Recei?ed 25 March 1999 and in re?ised form 8 September 1999)
We present a general regression-based method for mapping quantitative trait loci (QTL) by
combining different populations derived from diallel designs. The model expresses, at any map
position, the phenotypic value of each individual as a function of the specific-mean of the
population to which the individual belongs, the additive and dominance effects of the alleles
carried by the parents of that population and the probabilities of QTL genotypes conditional on
those of neighbouring markers. Standard linear model procedures (ordinary or iteratively
reweighted least-squares) are used for estimation and test of the parameters.
Most methods for mapping quantitative trait loci
(QTL) are designated to handle a single cross between
two inbred or outbred parents. Combining infor-
mation from multiple crosses has proved to be a more
1996; Xu, 1996; Xie et al., 1998; Xu, 1998); it
increases the chance that polymorphic alleles are
present in the parental gene pool and allows the
estimation of QTL effects and positions over a larger
set of lines. Using multiple families of crosses also
the detection of QTLs which are undetectable in a
single-line cross, where the two parents could be fixed
for the same allele at a particular QTL.
We have considered QTL mapping using multiple
crossing diallel designs between a set of inbred lines
(Rebaı? & Goffinet 1993, 1996; Rebaı? et al. 1994,
the North Carolina design III, wherein the F2 from
two inbred lines are backcrossed to both parental
lines. Xu (1996) has developed methods for QTL
mapping using four-way crosses. Muranty (1996)
studied the power of different mating designs between
* Corresponding author: Centre of Biotechnology of Sfax, Lab-
oratory of Plant Protection and Transformation, B.P. ‘K’, 3038
outbreds for QTL detection using single-marker
methods. Xu (1998) described and compared two
strategies (fixed and random-model strategies) of
combining data from multiple families of line crosses.
Recently, Xie et al. (1998) have developed a new
approach based on an identity by descent variance
component method which allows QTL mapping by
combining different line crosses.
Regression-mapping with multiple markers has
proved to be a powerful and robust method for QTL
mapping in classical populations derived from
biparental crosses between inbred or outbred lines
(see Haley & Knott, 1992; Haley et al., 1994; Rebaı?,
1997). Its simplicity of generalization and implemen-
tation and its computation efficiency relative to
computer-intensive likelihood-based methods have
made it an attractive approach to QTL mapping in
In our previous work, we generalized and used this
method to analyse multiple populations derived from
complete half-diallel crosses among inbred lines. In
this paper, we give a further generalization of the
method for the analysis of incomplete diallels among
inbred and outbred lines.
2. Model and methods
In this section we first consider a complete half-diallel
cross between l parental inbred lines, where p?
l(1?1)?2 F2 populations are derived from the F1
A. Rebaı? and B. Goffinet
hybrids. The generalization to other population types
such as recombinant inbred lines, double haploids
evaluated per se or by testcross is straightforward. We
then describe in Section 3 the application of the
method to incomplete diallel crosses and diallels of
(i) The mapping model
Consider that each parental line Li (i?1…l) has a
genotypes and l(l?1) 2 heterozygotes (g?l(l?1)?2
genotypes). For any individual k of the F2 population
derived from the cross Li?Lj and having phenotypic
value yijk, the model could be written, at any given
map position, as:
where µgis a cross-specific mean, aiis the additive
effect of allele Qi and dijthe dominance effect between
alleles Qi and Qj (so that genotypes QiQi and QiQj
have genotypic values of 2aiand ai?aj?dij, re-
spectively), and eijkare the residuals assumed to be
that individual k has genotype QiQi for the putative
QTL conditional on its observed genotype for the
markers GM and is a function of the distances
between markers and the position of the QTL. The
expressions of these probabilities are easy to obtain
for two flanking markers (e.g. Rebaı? et al., 1994). If l
parents are involved, only (l?1) additive parameters
are estimable and a constraint on the aivalues should
be used (e.g. Σiai?0) while all dominance parameters
are estimable. Thus the total number of estimable
parameter is q?(l?1)(l?2)?2.
In matrix notation, model (1) could be expressed
where Y is the n?1 vector of phenotypic values of n
individuals from p different populations, βois a p?1
vector of cross-specific means, Xois a n?p matrix
whose (ij)th element is 0 or 1 according to whether
or not individual i (i?1…n) belongs to population
j (j?1…p), βqis a q?1 vector of QTL effects, Xqis
a n?q matrix whose elements are linear combinations
of the probabilities of QTL genotypes conditional on
those of neighbouring markers and e is a n?1 vector
of residuals with known variance matrix Var(e)?σ?
I (I being the identity matrix). β?q(? stands for the
transpose) could be partitioned as [βa?βd]?, where βa
parameters. X and β could thus be partitioned as:
Note that only Xqneeds to be considered at each
genome position (Xois built once).
(ii) Construction of Xq
Xqcould be obtained by a product of two matrices
C is a g?q matrix of constants expressing the con-
straints on the parameters. g is the number of all
possible genotypes at the QTL (g?l(l?1)?2). The
element of line i and column j of P is the probability
that individual i has the jth genotype at the QTL (for
a given position) conditional on its genotype at neigh-
bouring marker loci.
(iii) Computation of the probability matrix, P
At any given genome position, elements of P are
p(ij)?Pr(GQ(i)?j?GM(i)), where GQ(i) is the QTL
genotype (j denotes the jth possible QTL genotype)
and GM(i) is the vector of genotypes of individual i for
all markers of the chromosome. For computation of
p(ij) only a subset of markers is used (those which
provide information on the QTL genotype). The
number of markers used in GM(i) would vary among
individuals and among positions for the same in-
dividual. The first step in computing p(ij) is thus to
extract the vector of informative markers from GM(i).
Let us denote this vector by gm(i); gm(i) thus contains
genotypes of markers which are informative for
individual i (not missing). If markers are co-dominant
then only the two informative markers (having no
missing data) flanking the QTL position are useful. If
a dominant marker is encountered (at left or right of
the genome position under study) other markers are
included in the analysis. Limiting the number of
informative markers to be used on each side of the
as little gain of information is expected from using
additional markers. Two closely linked (less than
5 cM apart) dominant markers linked in repulsion (on
the same side of the QTL) have the same information
content as a single co-dominant marker (Plomion et
(iv) Estimation and test of QTL effects
Applying model (2) at each genome position, par-
ameter estimates and tests are computed using
QTL mapping in diallel designs
standard linear model procedures (ordinary least-
squares and F-test). Note that taking Var(e)?σ?R,
where R is a diagonal matrix whose elements are
functions of QTL probabilities and QTL parameters,
increases the accuracy of estimates when the QTL
effects are large and?or marker density is weak (Xu,
1998). In this case, an iteratively reweighted least-
squares algorithm should be used.
Given estimates of aiand dij, parameters, additive
and dominance QTL variances, denoted respectively
as σ?aand σ?d, could be estimated by the expressions
(see also Xu, 1998):
where X−?(X?X)−1X?. For complete half-diallel
crosses with l parents these expressions simplify to
σ?a?(Σia?i)?(l?1) and σ?d?2(Σd?ij)?l(l?1). One can
calculate an estimate of the variance due to the QTL
as σ?q?σ?a?σ?dand thus the part of total variance
explained by the QTL as r??σ?q?(σ?q?σ?). A domi-
nance ratio at the QTL could also be calculated as
σd?σa. In all the above expressions, parameters are
replaced by their estimated values and the estimates so
obtained are asymptotically unbiased.
In our model it is assumed a priori that the QTL is
but it is possible to have a rough idea a posteriori of
the actual number of alleles segregating for the QTL.
This can be achieved by pairwise comparisons of
additive effects at the QTL (using t-tests) since the
sampling variances of these effects could be obtained
3. Application to incomplete diallels and outbred
(i) Application to diallels of inbred lines with missing
When the number of parents used in the diallel is large
deriving and evaluating all possible crosses could
disconnected small diallels or a partial diallel by
choosing some crosses of particular interest. In this
latter case the number of populations p is less than
l(l?1)?2 and some of the genotypes are not observed.
One can show that if p?l?1 and there is no common
parent between families then the design could no
longer be considered as a diallel but QTL mapping
can still be done with a different model by combining
information from all populations (see for example Xu,
1998); however, with such an approach QTL effects
are estimated using within-family information and
would be similar to those obtained at the same map
position from each family analysed separately (simu-
lation results not shown).
If p?l?1 then there are least two populations with
a common parent. In this case, (l?1) additive
parameters of the QTL are estimable (with the usual
constraint Σai?0) but only p dominance parameters
are estimable (one for each observed heterozygous
genotype). However, this condition does not necess-
arily imply that the QTL analysis using the method
described above gives parameter estimates which are
more accurate (having a smaller sampling variance)
than the estimates obtained from an analysis where
the populations are considered as independent ones.
One can show that this is true if and only if the two-
way cross classification represented by the diallel is
connected in the sense of connectedness in N-ways
cross classifications (Weeks & Williams, 1964). Using
the algorithm proposed by Weeks & Williams (1964)
it is possible to check the connectedness of the diallel
design and, if it is not connected, to find all connected
subsets. These sub-diallels could therefore be analysed
independently by the method described above.
Note that the sampling error of the additive effect
of any QTL allele depends on the number of
replications of the parent carrying this allele. Parents
involved in more crosses will have their allele effects
estimated with better precision (simulation results not
(ii) Application to outbred populations
Full-sib families derived from diallel crosses among
outbred lines assuming known linkage phases of the
markers could be analysed using the model described
above with some modifications. In fact, for a full-sib
family, there are four types of markers depending on
the number of alleles differing between the two
parents; markers could be segregating for four, three
or two alleles. In this latter case, a cross between two
parents could be of three different types – aa?ab or
ab?aa (backcross-like) or ab?ab (F2-like) – where a
and b designate two different alleles. For three and
four alleles, the cross is of type ab?ac and ab?cd,
respectively and there are four different genotypes in
a full-sib family from such crosses. To calculate the
probabilities of possible QTL genotypes in outbred
crosses we also need information on the linkage phase
of the markers which allows the haplotypes of the
parents to be deduced. These linkage phases could be
inferred from the marker data.
In the QTL mapping model we assume a priori that
the QTL is fully informative among the parents (each
outbred parent has two different alleles which are also
different from those of other parents). Let us denote
as bi and ci the alleles carried by the parent Li. With
A. Rebaı? and B. Goffinet
l parents and for a complete half-diallel, the number
of possible genotypes for the QTL is thus g?4p?
2l(l?1). TheestimableQTL parameters are:l additive
parameters (one for each parent Li) and l(l?1)?2
dominance parameters (one for each population). We
denote as aithe additive effect of alleles of parent Li
so that alleles bi and ci contribute to the genotypic
value with ?aiand ?ai, respectively. In a full-sib
family from the cross Li?Lj the genotypic values of
the four possible genotypes are thus:
In this case, matrix P will be constructed at the within-
family level: the first four columns of P are relative to
the first family, the second four columns to the second
family, etc. For the computation of the probabilities
of possible genotypes at the QTL, markers are used
sequentially until a fully informative marker is
encountered or maximum information is reached.
Note that these probabilities depend on both marker
genotypes and linkage phases. For an incomplete
of possible genotypes at the QTL is 4p and the number
of estimable parameters is (l?p) (l additive and p
dominance parameters) whatever the number of
observed families. There is no necessary condition on
the values of l and p for the diallel to be analysed.
The algorithms and statistical procedures used for the
expressions of matrices C) and for parameter test and
estimation are described in detail in Rebaı? (1996). The
programs for analysis of diallel data are now available
in the MultiCrossQTL software (Rebaı? et al., 1997b).
The QTL mapping method described here is a general
approach for combining data from different crosses. It
could quite easily be adapted to other mating designs
or for independent populations (having no parents in
common) of different types. Data from different
experiments and locations may be combined as well.
The generalization of our approach for multiple
QTL mapping with cofactors (Jansen & Stam, 1994) is
theoretically easy but may encounter some difficulty
In fact, if fully informative markers are not available
one may be obliged to use markers which will be non-
informative in some crosses as cofactors. In this case,
for each non-informative marker in a given cross the
nearest informative marker will be used to compute
the probability of marker genotypes, thereby compli-
cating the analysis. An alternative is to use different
sets of marker cofactors for different families, but this
could increase substantially the number of model
parameters, especially when the number of families is
large, and may result in a significant loss in power.
Finally, it would be interesting to compare the
performance of our fixed-model strategy for diallel
analysis with the IBD-based random model approach
of Xie et al. (1998) and, in particular, to study their
relative power as the number of parents increases. Xu
(1998) has shown that the random model approach is
computationally superior to the fixed model when the
number of families is large, but the two strategies
perform equally well. We think that there could be a
critical number of parents above which a random
approach may become more powerful, but this will be
Cockerham, C. C. & Zeng, Z.-B. (1996). Design III with
marker loci. Genetics 143, 1437–1456.
Haley, C. S. & Knott, S. A. (1992). A simple regression
method for mapping quantitative trait loci in line crosses
using flanking markers. Heredity 69, 315–324.
Haley, C. S., Knott, S. A. & Elsen, J. M. (1994). Mapping
quantitative trait loci in crosses between outbred lines
using least squares. Genetics 136, 1195–1207.
Jansen, R. C. & Stam, P. (1994). High resolution of
quantitative traits into multiple loci via interval mapping.
Genetics 136, 1447–1455.
Muranty, H. (1996). Power of tests for quantitative trait loci
detection using full-sib families in different schemes.
Heredity 76, 156–165.
Plomion, C., Liu, B.-H., & O’Malley, A. E. (1995). Genetic
analysis using trans-dominant markers in F2 family.
Theoretical and Applied Genetics 93, 1083–1089.
Rebaı?, A. (1996). Rapport de spe?cification du logicel de
cartographie de QTL chez les plantes: MCQTL. Technical
report. Institut National de la Recherche Agronomique,
Unite? de Biome?trie et Intelligence Artificielle, Toulouse.
Rebaı?, A. (1997). Comparison of methods for regression
interval mapping in QTL analysis with non-normal traits.
Genetical Research 69, 69–74.
Rebaı?, A. & Goffinet, B. (1993). Power of tests for QTL
detection using replicated progenies derived from a diallel
cross. Theoretical and Applied Genetics 86, 1014–1022.
Rebaı?, A. & Goffinet, B. (1996). Correction: power of tests
for QTL detection using replicated progenies derived
from a diallel cross. Theoretical and Applied Genetics 92,
Rebaı?, A., Goffinet, B., Mangin, B. & Perret, D. (1994).
Detecting QTL with diallel schemes. In Biometrics in
Plant Breeding: Applications of Molecular Markers.
Proceedings of the Ninth Eucarpia Meeting, section Bio-
metrics in Plant Breeding (Jansen, J. & van Ooijen, J. W.,
editors), pp. 170–177, Wageningen, The Netherlands.
Rebaı?, A., Blanchard, P., Perret, D. & Vincourt, P. (1997a).
Mapping quantitative trait loci controlling silking date in
a diallel cross among four lines of maize. Theoretical and
Applied Genetics 95, 451–459.
Rebaı?, A., Jourjon, M. F., Goffinet, B. & Mangin, B.
(1997b). QTL mapping in multiple crossing designs using
MulticrossQTL. In Ad?ances in Biometrical Genetics,
QTL mapping in diallel designs
Proceedings of the Tenth Eucarpia Meeting, section
Biometrics in Plant Breeding, (Krajewskiand, P. &
Kaczmarek, Z., editors), pp. 225–229, Poznan, Poland.
Weeks, D. L. & Williams, D. R. (1964). A note on the
determination of connectedness in an N-way cross
classification. Technometrics 6, 319–324.
Xie, C., Gessler, D. G. & Xu, S. (1998). Combining different
line crosses for mapping quantitative trait loci using the
identical by descent-based variance component method.
Genetics 149, 1139–1146.
Xu, S. (1996). Mapping quantitative trait loci using four-
way crosses. Genetical Research 68, 175–181.
Xu, S. (1998). Mapping quantitative trait loci using multiple
families of line crosses. Genetics 148, 517–524.