Marker-Assisted Introgression of Major QTLs for Grain Yield Under Drought in Rice - Translational Genomics for Crop Breeding: Abiotic Stress, Yield and Quality - page 60

Fig. 4.6. Marker-assisted backcrossing strategy for large-scale pyramiding of drought yield
QTLs with genes/QTLs for tolerance of biotic stresses such as bacterial leaf blight ( Xa21 ), gall
midge ( Gm8 ), blast ( Pi9 ), and submergence ( SUB1 ). FS- foreground selection, RS- recombinant
selection, BS- background selection.

are linked to QTLs affecting the trait of interest
and using them to trace the presence of QTLs in a
different generation of a backcross program lead-
ing to NILs with QTLs affecting the trait of inter-
est. Recent advances in high-throughput marker
technology have led to a significant increase in
the number of markers available in rice. Sin-
gle nucleotide polymorphisms (SNPs) in partic-
ular provide for a very elaborate coverage of the
rice genome. Moreover, high-throughput facili-
ties for SNP genotyping have markedly increased
the efficiency and reduced the cost of genotyp-
ing. These facilities can now be exploited for
rapid population improvement programs for a
diverse set of traits. Two approaches that enable
the use of markers for population improvement
are marker-assisted recurrent selection (MARS)
and genomic selection (GS). MARS refers to
the improvement of an F 2 population by one
cycle of marker-assisted selection (i.e., based on

phenotypic data and marker scores), followed
by three cycles of marker-based selection that
is based on marker scores only (Bernardo and
Charcosset 2006). The marker scores are usually
determined from 20 to 35 markers that show sig-
nificant association with one or more traits of
interest in a multiple-regression model (Edwards
and Johnson 1994; Koebner 2003). In contrast,
GS refers to marker-based selection without sig-
nificance testing and without identifying a subset
of markers associated with the trait (Meuwis-
sen et al. 2001). In this technique, the lines are
genotyped with markers spread evenly across
the genome. The effects on the quantitative trait
(i.e., breeding values) of all markers are fitted
as random effects in a linear model. Trait val-
ues are then predicted as the sum of an individ-
ual's breeding values across all marker loci, and
selection is subsequently based on these genome-
wide predictions (Bernardo and Yu 2007). A high