fpls-12-745379 January 3, 2022 Time: 12:58 # 1
ORIGINAL RESEARCH
published: 07 January 2022
doi: 10.3389/fpls.2021.745379
Genomic Selection for Wheat Blast in
a Diversity Panel, Breeding Panel and
Full-Sibs Panel
Philomin Juliana1, Xinyao He2, Felix Marza3, Rabiul Islam4, Babul Anwar4,
Jesse Poland5, Sandesh Shrestha5, Gyanendra P. Singh6, Aakash Chawade7,
Arun K. Joshi1,8, Ravi P. Singh2* and Pawan K. Singh2*
1 Borlaug Institute for South Asia (BISA), Ludhiana, India, 2 International Maize and Wheat Improvement Center (CIMMYT),
Mexico, Mexico, 3 Instituto Nacional de Innovación Agropecuaria y Forestal (INIAF), La Paz, Bolivia, 4 Bangladesh Wheat
and Maize Research Institute (BWMRI), Dinajpur, Bangladesh, 5 Department of Plant Pathology, Wheat Genetics Resource
Center, Kansas State University, Manhattan, KS, United States, 6 Indian Council of Agricultural Research (ICAR)-Indian
Institute of Wheat and Barley Research, Karnal, India, 7 Department of Plant Breeding, Swedish University of Agricultural
Sciences, Alnarp, Sweden, 8 CIMMYT-India, New Delhi, India
Edited by:
Christina Cowger, Wheat blast is an emerging threat to wheat production, due to its recent migration to
United States Department South Asia and Sub-Saharan Africa. Because genomic selection (GS) has emerged
of Agriculture (USDA), United States as a promising breeding strategy, the key objective of this study was to evaluate it
Reviewed by: for wheat blast phenotyped at precision phenotyping platforms in Quirusillas (Bolivia),
Luxiang Liu,
Institute of Crop Sciences, Chinese Okinawa (Bolivia) and Jashore (Bangladesh) using three panels: (i) a diversity panel
Academy of Agricultural Sciences comprising 172 diverse spring wheat genotypes, (ii) a breeding panel comprising 248
(CAAS), China
Thomas Miedaner, elite breeding lines, and (iii) a full-sibs panel comprising 298 full-sibs. We evaluated two
University of Hohenheim, Germany genomic prediction models (the genomic best linear unbiased prediction or GBLUP
Tofazzal Islam, model and the Bayes B model) and compared the genomic prediction accuracies
Bangabandhu Sheikh Mujibur
Rahman Agricultural University, with accuracies from a fixed effects model (with selected blast-associated markers
Bangladesh as fixed effects), a GBLUP + fixed effects model and a pedigree relationships-based
*Correspondence: model (ABLUP). On average, across all the panels and environments analyzed, the
Ravi P. Singh
R.Singh@cgiar.org GBLUP + fixed effects model (0.63 ± 0.13) and the fixed effects model (0.62 ±
Pawan K. Singh 0.13) gave the highest prediction accuracies, followed by the Bayes B (0.59 ± 0.11),
Pk.Singh@cgiar.org GBLUP (0.55 ± 0.1), and ABLUP (0.48 ± 0.06) models. The high prediction accuracies
Specialty section: from the fixed effects model resulted from the markers tagging the 2NS translocation
This article was submitted to that had a large effect on blast in all the panels. This implies that in environments
Plant Breeding, where the 2NS translocation-based blast resistance is effective, genotyping one to
a section of the journal
Frontiers in Plant Science few markers tagging the translocation is sufficient to predict the blast response and
Received: 22 July 2021 genome-wide markers may not be needed. We also observed that marker-assisted
Accepted: 09 November 2021 selection (MAS) based on a few blast-associated markers outperformed GS as it
Published: 07 January 2022
selected the highest mean percentage (88.5%) of lines also selected by phenotypic
Citation:
Juliana P, He X, Marza F, Islam R, selection and discarded the highest mean percentage of lines (91.8%) also discarded
Anwar B, Poland J, Shrestha S, by phenotypic selection, across all panels. In conclusion, while this study demonstrates
Singh GP, Chawade A, Joshi AK, that MAS might be a powerful strategy to select for the 2NS translocation-based blast
Singh RP and Singh PK (2022)
Genomic Selection for Wheat Blast resistance, we emphasize that further efforts to use genomic tools to identify non-2NS
in a Diversity Panel, Breeding Panel translocation-based blast resistance are critical.
and Full-Sibs Panel.
Front. Plant Sci. 12:745379. Keywords: wheat, blast disease, genomic selection (GS), marker-assisted selection, pedigree selection,
doi: 10.3389/fpls.2021.745379 genotyping-by sequencing, Magnaporthe oryzae
Frontiers in Plant Science | www.frontiersin.org 1 January 2022 | Volume 12 | Article 745379
fpls-12-745379 January 3, 2022 Time: 12:58 # 2
Juliana et al. Genomic Selection for Wheat Blast
INTRODUCTION wheat blast resistance genes including Rmg2, Rmg3, Rmg7, Rmg8,
and RmgGR119, only the genes Rmg8 and RmgGR119 are known
An emerging threat to wheat production that has the potential to to be effective against several recent MoT isolates (Zhan et al.,
cause substantial yield losses is the disease blast (Kohli et al., 2011; 2008; Anh et al., 2015, 2018; Tagle et al., 2015; Cruz and
Islam et al., 2016; Chowdhury et al., 2017; Cruz and Valent, 2017; Valent, 2017; Wang S. et al., 2018). Besides these genes, the
Sadat and Choi, 2017; Singh et al., 2021), caused by the fungus 2NS translocation from the wild species, Aegilops ventricosa has
Magnaporthe oryzae pathotype Triticum Catt. (MoT) (anamorph been reported to confer a consistent and strong effect on blast
Pyricularia oryzae Cavara) (Couch and Kohn, 2002; Tosa and resistance in several studies, although the resistance is sometimes
Chuma, 2014; Zhang et al., 2016). The disease primarily affects background dependent and partial (Cruz et al., 2016b; Juliana
the spikes which become partially or fully bleached, resulting in et al., 2019, 2020a; He et al., 2020, 2021; Ferreira et al., 2021; Wu
inferior quality of grains which are small, shriveled and have low et al., 2021).
test weight (Goulart et al., 2007; Urashima et al., 2009; Cruz and Breeding for wheat blast resistant genotypes first involves
Valent, 2017). First identified in 1985 in Brazil (Igarashi, 1986), screening to find resistant germplasm and then identifying
the disease spread to the major Brazilian wheat growing areas resistance genes. However, wheat breeding programs globally are
(Goulart et al., 1990; Igarashi, 1990; Picinini and Fernandes, 1990; constrained in their ability to screen a large number of lines for
Dos Anjos et al., 1996), and then moved to Bolivia, Paraguay blast resistance, as phenotyping can only be done in the blast hot-
and Argentina in 1996, 2002, and 2007, respectively (Barea and spot locations and there is a limitation to the number of lines that
Toledo, 1996; Viedma and Morel, 2002; Cabrera and Gutiérrez, can be handled, unless their phenotyping capacity is expanded.
2007; Perelló et al., 2015). While this poses a huge challenge to accelerate development
The first intercontinental jump of the MoT pathogen from of blast resistant wheat varieties, it is an excellent case for the
South America to Asia was reported in 2016, when there was application of genomic selection (GS), an approach that was
a blast outbreak in Bangladesh most likely caused by the South advocated to change the role of phenotyping in breeding (Heffner
American lineage of MoT via wheat importation (Islam et al., et al., 2009). Using GS, breeders can eliminate phenotyping
2016; Malaker et al., 2016; Ceresini et al., 2018). In addition, and select genotypes based on their genomic-estimated breeding
the warm and humid climate at heading time during that year values (GEBVs) for traits, that are obtained from genome-wide
was also a significant driver of the epidemic, as both high markers (Meuwissen et al., 2001). In GS, a “training population”
temperatures (between 25 and 30◦C) and long wetting periods that includes lines that have been genotyped and phenotyped
favor blast development (Cardoso et al., 2008; Islam et al., 2019). for the trait of interest is used to train prediction models that
Another major intercontinental jump of the MoT pathogen are then used to obtain the GEBVs of individuals (also known
to Africa was recently reported, when blast was observed in as “selection candidates” or “testing population”) that have been
the Muchinga province of Zambia during the 2017–2018 rainy only genotyped. While GS has proved to be effective in predicting
season (Tembo et al., 2020). Furthermore, about seven million quantitative disease resistance (Ornella et al., 2012; Rutkoski
hectares of wheat growing regions in India, Pakistan and et al., 2014; Juliana et al., 2019), it also has the potential to increase
Bangladesh and some states in the United States (Louisiana, the accuracy of selection, reduce cycle time and cost, thereby
Mississippi and Florida) were identified to be vulnerable to blast leading to an increase in gain from selection (Heffner et al., 2010;
outbreaks, given their similar favorable environmental conditions Voss-Fels et al., 2019).
(Cruz et al., 2016a; Mottaleb et al., 2018; Valent et al., 2021), Given the potential of GS for wheat blast, the key objective
indicating that further spread of the disease is possible. of this study was to evaluate it in the following panels, assuming
Wheat blast management approaches like the use of that a subset or half of them were phenotyped: (a) Diversity
fungicides, planting time alteration and discontinuation of wheat panel comprising diverse spring wheat lines and varieties that
cultivation in disease-prone regions by declaring a wheat holiday were developed over several years by the International Maize
have only been partly successful in combating the disease and Wheat Improvement Centre (CIMMYT) and South Asia
(Mottaleb et al., 2019b; Roy et al., 2021). This is because of partners, which is useful to understand if GS can be applied
limitations such as inefficient control with fungicides when the to select for blast resistance in unrelated lines or any set of
disease pressure is high, inability of poor farmers to afford existing historic germplasm. (b) Breeding panel comprising elite
fungicides, development of resistance to some fungicide classes lines from CIMMYT’s international nurseries, which is useful
in MoT populations and challenges of finding the appropriate to understand if GS can be applied to select advanced breeding
profitable alternative wheat land use (Urashima et al., 2009; lines for blast resistance. (c) Full-sibs panel comprising progenies
Kohli et al., 2011; Castroagudín et al., 2015; Cruz et al., 2015, from a cross between a resistant and a susceptible blast parent,
2019; Coelho et al., 2016; Cruz and Valent, 2017; Mottaleb which is useful to understand if selection for blast is effective
et al., 2019a,b). Hence, the most sustainable, cost-effective and within families, i.e., among sister lines in biparental populations.
farmer-friendly approach to wheat blast control is developing and The other main objectives of this study were to:
deploying blast resistant wheat varieties (Cruz and Valent, 2017).
Genetic resistance to wheat blast is known to follow the gene- (i) compare genomic prediction accuracies from the genomic
for-gene interaction model in the seedling stage (Takabayashi best linear unbiased prediction (GBLUP) model that
et al., 2002), while field resistance is also known to be quantitative utilizes the genomic relationships between lines (de
(Goddard et al., 2020; He et al., 2021). Among the five reported los Campos et al., 2013; Habier et al., 2013) and the
Frontiers in Plant Science | www.frontiersin.org 2 January 2022 | Volume 12 | Article 745379
fpls-12-745379 January 3, 2022 Time: 12:58 # 3
Juliana et al. Genomic Selection for Wheat Blast
Bayes B model that utilizes the estimated marker effects (ii) Okinawa, Bolivia (17◦13′S 62◦53′W) during the 2018 crop
(Meuwissen et al., 2001) to generate GEBVs. cycle (May to September) in two planting dates and
(ii) compare genomic prediction accuracies from both the the datasets are referred to as Okinawa 2018 FP and
genomic prediction models (GBLUP and Bayes B) with Okinawa 2018 SP.
prediction accuracies from a fixed effects model, where a (iii) Jashore, Bangladesh (23◦10′N 89◦10′E) during the 2017–
genome-wide association analysis for blast is first done in 2018 crop cycle (December to April) in two different
the training set, followed by selection of the best model planting dates and the datasets are referred to as Jashore
(when adding a marker to the model no longer increases 2018 FP and Jashore 2018 SP.
the prediction accuracy) and use of the selected marker(s)
to estimate the breeding values, referred to as the estimated Breeding Panel
breeding values (EBVs). The breeding panel comprised 248 lines from CIMMYT’s
(iii) compare prediction accuracies from the GBLUP model international nurseries that included subsets of lines from the
and the fixed effects model to the accuracies from 50th International Bread Wheat Screening Nursery (IBWSN,
the combined GBLUP and the fixed effects model 119 lines) and the 35th Semi-Arid Wheat Screening Nursery
(GBLUP+ fixed effects). (SAWSN, 129 lines). The IBWSNs and SAWSNs comprise
(iv) compare genomic prediction accuracies with pedigree- advanced breeding lines developed by CIMMYT’s global wheat
based prediction accuracies, where pedigree-based program using the selected bulk-breeding scheme that are
relationships between the lines is used to obtain the targeted to the irrigated and drought-prone target environments,
EBVs, in a pedigree (additive)-best linear unbiased respectively and are CIMMYT’s primary vehicles of germplasm
prediction model (ABLUP). dissemination globally (Rajaram et al., 1993; van Ginkel and
(v) compare selections made from the blast phenotypes Rajaram, 1993). From the set of 269 lines from the 50th IBWSN
(phenotypic selection, PS) with selections using the EBVs and 265 lines from the 35th SAWSN (Juliana et al., 2020a), subsets
from the different models to understand what percentage of lines were chosen after filtering out a large number of lines
of lines that are selected and discarded by PS, overlap with that had across-environment blast best linear unbiased estimates
the breeding values-based selections. (BLUEs) of 0, and only some of those lines were retained to avoid
(vi) test the hypothesis that GS would perform better than a large number of lines with a blast index of 0 in the training and
the selections based on EBVs from a fixed-effects model prediction populations. Similarly, only the environments where
(which can be considered similar to marker-assisted more than half the entries did not have a blast index of 0 were
selection, MAS) and the pedigree relationships-based chosen. The selected environments where the breeding panel was
model (pedigree selection). phenotyped for blast included:
(vii) compare prediction accuracies in subsets of lines with and
without the 2NS translocation in the three panels using the (i) Quirusillas during the 2017–2018 crop cycle (December
GBLUP, Bayes B, fixed effects, GBLUP + fixed effects and to April) in the FP date and in the 2018–2019 crop
ABLUP models. cycle in two different planting dates and the datasets are
referred to as: Quirusillas 2018 FP, Quirusillas 2019 FP and
Quirusillas 2019 SP.
(ii) Okinawa during the 2018 crop cycle (May to September)
MATERIALS AND METHODS where only the second planting was chosen (Okinawa 2018
SP), due to the high number of resistant lines in the FP.
Panels, Blast Evaluation Sites, Crop
Cycles, and Planting Time Full-Sibs Panel
The full-sibs panel comprised 298 full-sibs or F2:7 recombinant
Diversity Panel inbred lines that were obtained by single seed descent from a
The diversity panel comprised 172 diverse spring wheat cross between a resistant female parent Caninde#1 (with the 2NS
genotypes including lines developed by CIMMYT and varieties translocation) and a susceptible male parent Alondra (without
released in South Asia (India, Bangladesh, and Nepal), some of the 2NS translocation), as described in He et al. (2020). The full-
which were directly introduced from CIMMYT. The diversity sibs panel was phenotyped for blast in two planting dates in the
panel was phenotyped for blast in two planting dates that following sites and crop cycles:
were about 14 days apart, indicated as first planting (FP) and
second planting (SP) in the following blast precision phenotyping (i) Quirusillas during the 2017–2018 and 2018–2019 crop
platforms and crop cycles: cycles (December to April) in two different planting
dates and the datasets are referred to as: Quirusillas
(i) Quirusillas, Bolivia (18◦20′S 63◦57′W) during the 2017– 2018 FP, Quirusillas 2018 SP, Quirusillas 2019 FP and
2018 and 2018–2019 crop cycles (December to April) in Quirusillas 2019 SP.
two different planting dates and the datasets are referred to (ii) Okinawa during the 2018 and 2019 crop cycles (May to
by the site followed by the harvest year and planting time September) in two planting dates and the datasets are
as: Quirusillas 2018 FP, Quirusillas 2018 SP, Quirusillas referred to as Okinawa 2018 FP, Okinawa 2018 SP, Okinawa
2019 FP and Quirusillas 2019 SP. 2019 FP and Okinawa 2019 SP.
Frontiers in Plant Science | www.frontiersin.org 3 January 2022 | Volume 12 | Article 745379
fpls-12-745379 January 3, 2022 Time: 12:58 # 4
Juliana et al. Genomic Selection for Wheat Blast
(iii) Jashore during the 2017–2018 and 2018–2019 crop cycles (Gilmour et al., 1995) in the “R” package “heritability”
(December to April) in two planting dates and the datasets (Kruijer et al., 2015).
are referred to as Jashore 2018 FP, Jashore 2018 SP, Jashore
2019 FP and Jashore 2019 SP. Genotyping
The diversity panel was genotyped for genome-wide markers
Blast Phenotyping—Field Experimental Design, using the Illumina Infinium 15K BeadChip (TraitGenetics,
Inoculation, Evaluation, and Analyses Germany) and four sequence tagged site (STS) markers
In all the three sites, the lines were planted in double rows associated with the Yr17 gene in the 2NS translocation namely:
each of 1-m length with 20-cm spacing in between them. Blast Ventriup (Helguera et al., 2003), WGGB156 and WGGB159
inoculation in Quirusillas and Okinawa was done using isolates (Wang Y. et al., 2018) and cslVrgal3 (Seah et al., 2001; He
QUI1505, QUI1601, QUI1612, OKI1503, and OKI1704 and et al., 2021). The breeding panel was genotyped for genome-wide
in Jashore it was done using isolates BHO17001, MEH17003, markers using the genotyping-by-sequencing (GBS) platform
GOP17001.2, RAJ17001, CHU16001.3, and JES16001, all of (Poland et al., 2012) and the TASSEL (Trait Analysis by
which were collected locally and exhibited high pathogenesis. aSSociation Evolution and Linkage) version 5 GBS pipeline
Inoculum was prepared according to He et al. (2020) by (Glaubitz et al., 2014) was used to call the marker polymorphisms.
culturing the MoT isolates on oatmeal agar medium. Inoculum Marker polymorphisms discovery, alignment to the reference
concentration was adjusted to 80,000 spores/mL and applied genome assembly (RefSeq v1.0) of Chinese Spring (IWGSC,
using a backpack sprayer at anthesis, followed by a second 2018) and tag filtering were done as described in Juliana
inoculation 2 days later, in all the environments. et al. (2020a). The full-sibs panel was genotyped for genome-
Disease development after inoculation was favored using a wide markers using the DArTseq platform (Genetic Analysis
misting system that was set up to provide 10 min of misting each Service for Agriculture, CIMMYT, Mexico), the four STS markers
hour, between 8 a.m. and 7 p.m. in the Bolivian sites and between mentioned above and also another marker IWB11136 tagging the
9 a.m. to 5 p.m. in Jashore. In addition to the panel lines, local 2NS translocation (Xue et al., 2018). The genome-wide markers
checks were also planted and evaluated for blast, which included in each panel and STS markers were filtered for those with less
resistant check Urubo and susceptible check Atlax in Bolivia than 60% missing data, greater than 10% minor allele frequency
and resistant check BARI Gom 33 (Hossain et al., 2019) and and less than 10% heterozygosity resulting in 13,427 markers in
susceptible check BARI Gom 26 in Jashore. Evaluation of wheat the diversity panel, 8,072 markers in the breeding panel and 2,489
blast was done 21 days post the first inoculation on 10 spikes markers in the full-sibs panel. Marker imputation in all the panels
marked at anthesis, where the total number of spikelets and those was done using the linkage disequilibrium k-nearest neighbor
infected were counted. Wheat blast index was obtained using the genotype imputation method (Money et al., 2015) in TASSEL
formula: index = incidence (proportion of spikes that had blast version 5 (Bradbury et al., 2007).
infection)× severity (average percentage of infected spikelets).
The BLUEs for blast in each of the panels were calculated Blast Prediction
using the ASREML statistical package (Gilmour, 1997) using the Blast prediction in all the panels was done using a twofold cross-
following mixed model: validation approach, where each of the panels was divided into
two random folds and one-half of the lines was used to predict
yij = µ+ gi + ej + εij (1) the breeding values of the other half of the lines for blast within
each panel. We have only evaluated twofold cross-validations,
where yij is the observed blast index, µ is the overall mean, because across the panels, 15.8–62.5% of the lines had a blast
gi is the fixed effect of the genotype, ej is the random effect of index of zero and dividing them into smaller folds might result in
the environment (site-year-planting time) that was independent some random folds having most of the lines with a blast index of
and identically distributed (IID) (ej ∼ N (0, σ2
e )), and εij is zero. The sampling of the random folds was iterated 10 times, the
the residual with IID (εij ∼ N (0, σ2
ε )). Analysis of the blast prediction accuracy was calculated as the Pearson’s correlation
indices in the different panels and environments was done and between the observed blast index values and the breeding values
the mean, standard deviation, median, minimum and maximum in each iteration and the mean prediction accuracy across the 10
blast indices were obtained in all the datasets. Visualization of all iterations was obtained for each of the datasets in the different
the results in this study was done using the “R” package “ggplot2” panels using the following models:
(Wickham, 2009). The narrow-sense heritabilities for blast across
the different environments in each panel were obtained using the (i) Fixed effects model
formula: For the fixed effects model implemented in “R,” a stepwise
h2 σ2
A
=
σ2 (2) least-squares approach was used which involved the following
A + σ2
ε steps:
where σ2
A was the additive genetic variance among the lines • Identification of markers significantly associated with blast
calculated using markers and σ2
ε is the error variance. The in the training set using a genome-wide association analysis
heritabilities, genetic and error variances were obtained using the and calculation of marker p-values.
average information-restricted maximum likelihood algorithm • Ranking of markers according to their p-values.
Frontiers in Plant Science | www.frontiersin.org 4 January 2022 | Volume 12 | Article 745379
fpls-12-745379 January 3, 2022 Time: 12:58 # 5
Juliana et al. Genomic Selection for Wheat Blast
• Marker selection from the ranked markers was done with “BGLR” (Pérez and de los Campos, 2014), using the default prior
the following stepwise regression model: parameters and 10,000 iterations, while the first 1,000 iterations
were discarded as burn-in.
y = 1nµ+ Xiβi . . . Xjβj + ε (3)
(v) Pedigree-best linear unbiased prediction (ABLUP)
where y was the blast phenotype, µ was the mean, βiand βj were
the effects of the ith and jth marker, and Xi and Xj were the The ABLUP was a modified version of the GBLUP that was
ith and jth marker’s genotype matrix and ε was the error term. also implemented in the “R” package “BGLR,” where the marker-
Here, for each iteration i through j, we added a marker to the based genomic relationship matrix was replaced by the pedigree-
model, starting from the marker that had the lowest p-value. We based relationship matrix, that was calculated from the coefficient
then calculated the twofold cross validation accuracy within the of parentage and the pedigree tracing back to five generations.
training set after each iteration and selected the model that had j-1 To compare the prediction accuracies obtained from the
markers, when the accuracyj−1 was greater than the accuracyj. different models and to test if they were significantly different
from each other, we performed paired-t-tests using the “JMP”
• Estimation of marker effects was done from the selected statistical software1 and obtained the mean differences between
markers, and the effects were subsequently used for the prediction accuracies from the different model pairs in each
obtaining the EBVs of lines in the testing populations for panel. We also obtained the p-values to test their significance
blast resistance. at a threshold of 0.005 for three alternate hypotheses: the mean
(ii) Genomic-best linear unbiased prediction (GBLUP) prediction accuracy of one model is significantly greater or less
than the other model (two-tailed t-test), the mean prediction
The GBLUP model was fitted using the “R” package “rrBLUP” accuracy of one model is significantly greater than the other
(Endelman, 2011) and can be represented by the following mixed model (one-tailed t-test) and the mean prediction accuracy of
model: one model is significantly lesser than the other model (one-tailed
y = µ1+Zgu+ε (4) t-test).
where y was the vector of blast indices, µ was the mean, u Comparison of Genomic Selection With
represented the additive genetic effects, Z was the design matrix Marker-Assisted Selection and Pedigree-Based
for the random effects and ε was the error term. The joint Selection
distribution of u (the vector of additive genetic effects) was The BLUEs dataset in all the panels was used to select the most
assumed to be multivariate normal i.e., MN (0, Gσ2
g), where G resistant blast lines using the phenotypes (PS) and compared to
was the marker-based genomic relationship matrix calculated the following selections made using the EBVs for blast obtained
using the method of VanRaden (2008) [G = ZZ′/p, where Z from different models: (i) MAS using the EBVs obtained from
was the centered and standardized marker matrix and p was the the fixed effects model (ii) GS using the GEBVs obtained from
number of markers] and σ2
g was the genetic variance]. The joint the GBLUP (GS GBLUP) and Bayes B (GS Bayes B) models (iii)
distribution of ε (error term) was also assumed to be multivariate GS + MAS using the GEBVs obtained from the GBLUP + fixed
normal i.e., MN (0, Iσ2
e), where I was the identity matrix and σ2
e effects model (iv) pedigree selection using the EBVs obtained
was the residual variance. from the ABLUP model. For PS, we selected the lines with blast
(iii) Genomic-best linear unbiased prediction and fixed indices less than 10 in the BLUEs dataset for all the panels and
effects (GBLUP+ fixed effects) an equal number of lines were selected using the EBVs obtained
from the different models.
In the GBLUP + fixed effects model, in addition to modeling
the markers as random effects in the GBLUP model, some Blast Prediction in Subsets of Lines With and Without
loci were also modeled as fixed effects and the model can be the 2NS Translocation
represented as: Subsets of lines with and without the 2NS translocation were
obtained using consensus data from the STS markers tagging
y = 1nµ+ Xiβi . . . Xjβj + Zgu+ε (5) the 2NS translocation in the diversity and full-sibs panels and
using all the 2AS markers significantly associated with blast
where the terms are the same as described in (3) and (4). in the fixed effects model in the breeding panel. The lines
(iv) Bayes B where the presence or absence of the 2NS translocation could
not be determined using all the markers (because of missing
In the Bayes B model (Meuwissen et al., 2001), a mixture data or contrasting information from different markers) were
distribution prior is used, where some marker effects are excluded from predictions. Within the subsets of lines with and
assumed to be zero with probability, π (the markers linked without the 2NS translocation, blast prediction was done using
to regions of the genome that have no effect on the trait and twofold cross-validations with the fixed effects, GBLUP, Bayes B,
hence zero effect), and some marker effects are assumed to be GBLUP+ fixed effects and ABLUP models. The mean prediction
drawn from a scaled-t distribution with probability, 1-π (the accuracies obtained from the subsets with and without the 2NS
markers linked to regions of the genome that have an effect
on the trait). The Bayes B model was fitted in the “R” package 1www.jmp.com
Frontiers in Plant Science | www.frontiersin.org 5 January 2022 | Volume 12 | Article 745379
fpls-12-745379 January 3, 2022 Time: 12:58 # 6
Juliana et al. Genomic Selection for Wheat Blast
translocation in each of the panels were compared using paired- (259,187 bps, 0 cM), Kukri_c22599_114 (397,565 bps,
t-tests. 0 cM), Tdurum_contig11802_864 (2,478,927 bps, 0 cM),
Ventriup (3,965,255 bps, 0 cM), wsnp_Ku_c33374_42877546
(4,789,998 bps, 2.9 cM), Kukri_c31776_1621 (7,550,063 bps, 8.9
RESULTS cM), AX-94629608 (14,327,985 bps, 8.9 cM) and AX-94684111
(27,276,097 bps, 9.8cM), that were located between 259,187 and
Diversity Panel 27,276,097 bps on the Refseq v1.0 (IWGSC, 2018) and between 0
Statistical Analysis of Blast Indices in the Diversity and 9.8 cM on the Popseq map (Chapman et al., 2015).
Panel With all the models, the blast BLUEs had the highest
Statistical analysis of blast indices in the diversity panel prediction accuracies (0.61–0.85) in the diversity panel, that
(Supplementary Data 1 and Table 1) indicated that the mean were 34.2–45.5% higher than the mean prediction accuracies
blast indices were relatively higher in the Quirusillas 2019 FP of the individual environments. Considering all the models, we
(38.5 ± 35.1), Quirusillas 2018 FP (32 ± 25.5) and Okinawa observed that the mean prediction accuracy was the highest in
2018 SP (31.4 ± 22.9) datasets. The maximum blast index in Okinawa 2018 SP (0.66± 0.12) and the lowest in Jashore 2018 SP
the individual diversity panel datasets ranged between 48 and (0.44 ± 0.08) dataset. We observed that the mean differences in
100. We also observed that 23.3% (Jashore 2018 SP) to 43% prediction accuracies were not significant in the two-tailed t-tests
(Quirusillas 2018 SP) of the lines in the different environments at a threshold of 0.005 for the following model pairs:
had a blast index of zero. The phenotypic correlations between (i) GBLUP+ fixed effects and Bayes B: Mean difference = 0.04,
the blast indices in the two plantings were high in Quirusillas p-value = 0.21
2019 (0.7), while they were moderate in Okinawa 2018 (0.58), (ii) GBLUP + fixed effects and GBLUP: Mean
Quirusillas 2018 (0.56), and Jashore 2018 (0.46). Across the sites difference = 0.06, p-value = 0.09
of blast evaluation, we observed low to moderate correlations (iii) GBLUP + fixed effects and fixed effects: Mean
between the blast indices in Jashore and the Bolivian sites (ranged difference = 0.02, p-value = 0.03
between 0.27 and 0.53), while moderate to high correlations (iv) Bayes B and GBLUP: Mean difference = 0.01, p-value = 0.17
(ranged between 0.47 and 0.67) were observed between the (v) Fixed effects and Bayes B: Mean difference = 0.02, p-
blast indices in Okinawa and Quirusillas. The narrow-sense value = 0.57
heritability of blast across all the environments in the diversity (vi) Fixed effects and GBLUP: Mean difference = 0.03, p-
panel was 0.38 (σ2
A = 190.1 and σ2
ε = 308.2). value = 0.33
Prediction Accuracies for Blast in the Diversity Panel However, the prediction accuracies from all models were
The mean prediction accuracies for blast across the different significantly higher than the prediction accuracies from the
environments for all the lines in the diversity panel were: (i) ABLUP model at a threshold of 0.005 (p-values for the one-
0.63 ± 0.14 using the GBLUP + fixed effects model (ii) 0.60 ± sided t-test ranged from 4.03 × 10−6 to 2 × 10−3) and the
0.15 using the fixed effects model (iii) 0.58 ± 0.05 using the mean differences in prediction accuracies between the ABLUP
Bayes B model (iv) 0.57 ± 0.05 using the GBLUP model and and other models ranged from 0.12 to 0.18.
(v) 0.45 ± 0.03 using the ABLUP model (Figure 1). In the
fixed effects model, one or two markers on chromosome 2AS Phenotypic Selection vs. Estimated Breeding Values
(Supplementary Table 1) that were selected by association Based Selection for Blast in the Diversity Panel
analysis and stepwise regression were used as fixed effects For PS, we selected 52 lines (30.2%) with blast indices less than 10
in the different datasets (except the Jashore 2018 FP and SP in the BLUEs dataset and an equal number of lines using the EBVs
datasets). This included markers Tdurum_contig29983_490 from different models (Figure 2). Considering the GS+MAS and
MAS, we observed that among the 52 lines selected by PS, 94.2%
were also selected by these two selection methods. Similarly, 90.4
TABLE 1 | Statistical analysis of blast indices in the diversity panel with 172 lines. and 76.9% lines were selected by GS, using the GEBVs obtained
from the Bayes B and GBLUP models, respectively. Among the
Dataset Mean Standard Median Minimum Maximum
deviation 120 lines that were discarded by PS, 97.5, 97.5, 95.8, and 90%
were also discarded by GS + MAS, MAS, GS Bayes B and GS
Quirusillas 2018 FP 32.0 25.5 35.9 0 90.7 GBLUP, respectively. However, considering pedigree selection,
Quirusillas 2018 SP 22.3 24.1 11.0 0 77.0 we observed that 69.2% lines that were selected by PS were also
Quirusillas 2019 FP 38.5 35.1 36.9 0 100.0 selected by pedigree selection, while 86.7% lines that were not
Quirusillas 2019 SP 29.9 27.4 27.9 0 98.2 selected by PS were also not selected by pedigree selection.
Okinawa 2018 FP 21.8 20.5 18.3 0 76.4
Okinawa 2018 SP 31.4 22.9 38.7 0 72.7 Blast Distribution and Prediction Accuracies in
Jashore 2018 FP 11.2 12.3 8.9 0 48.0 Subsets of Lines With and Without the 2NS
Jashore 2018 SP 18.5 17.5 14.4 0 74.1 Translocation in the Diversity Panel
BLUEs 25.7 18.2 28.3 0 61.3 In the 53 diversity panel lines with the 2NS translocation, we
FP, First planting; SP, Second planting; BLUEs, Best linear unbiased estimates. observed that the mean blast index ranged between 1.5± 4.8 and
Frontiers in Plant Science | www.frontiersin.org 6 January 2022 | Volume 12 | Article 745379
fpls-12-745379 January 3, 2022 Time: 12:58 # 7
Juliana et al. Genomic Selection for Wheat Blast
FIGURE 1 | Twofold cross validation prediction accuracies for blast response in the diversity panel (172 lines) using the fixed effects (Fixed), genomic best linear
unbiased prediction (GBLUP), GBLUP and fixed effects (GBLUP + Fixed), Bayes B, and pedigree best linear unbiased prediction (ABLUP) models. FP refers to the
first planting, SP refers to the second planting and BLUEs refer to the best linear unbiased estimates of blast indices across the different environments.
FIGURE 2 | Comparison of phenotypic selection (PS) of the best linear unbiased estimates of blast indices across environments with: (i) marker assisted selection
(MAS) using the estimated breeding values (EBVs) obtained from the fixed effects model (Fixed) (ii) genomic selection (GS) using the genomic estimated breeding
values (GEBVs) obtained from the genomic best-linear unbiased prediction (GBLUP) and Bayes B models (iii) GS + MAS using the GEBVs obtained from the GBLUP
and fixed effects (GBLUP + Fixed) model and (iv) pedigree selection (PedS) using the EBVs obtained from the pedigree best linear unbiased prediction (ABLUP)
model in the diversity panel comprising 172 lines.
4.9± 7.3 in the different environments (Figure 3A). Similarly, in 2NS translocation, as several lines had a blast index of zero. The
the 119 diversity panel lines without the 2NS translocation, the markers used as fixed effects in the different datasets for the lines
mean blast index ranged between 14.3 ± 12.9 and 54.6 ± 29.8 with and without the 2NS translocation in the diversity panel are
in the different environments. The mean prediction accuracies given in Supplementary Tables 2, 3, respectively.
for blast across the different environments for the lines with We observed that the mean prediction accuracy across all the
the 2NS translocation in the diversity panel ranged between environments and models was significantly higher in the subset
0.04± 0.17 using the GBLUP+ fixed effects model and−0.03± of lines without the 2NS translocation compared to the subset
0.19 using the ABLUP model (Figure 3B). Similarly, for the of lines with the 2NS translocation (mean difference = 0.29, p-
lines without the 2NS translocation in the diversity panel, the value = 1.2 × 10−4). In the diversity panel lines where the 2NS
mean prediction accuracies ranged between 0.36 ± 0.18 using translocation was present, the mean prediction accuracy across
the Bayes B model and 0.27 ± 0.19 using the GBLUP + fixed all the models was the highest in Jashore 2018 SP (0.4± 0.06) and
effects model. The prediction accuracies could not be obtained the lowest in Okinawa 2018 FP (−0.16 ± 0.06). In the diversity
for some environments and models in the subset of lines with the panel lines where the 2NS translocation was absent, we observed
Frontiers in Plant Science | www.frontiersin.org 7 January 2022 | Volume 12 | Article 745379
fpls-12-745379 January 3, 2022 Time: 12:58 # 8
Juliana et al. Genomic Selection for Wheat Blast
FIGURE 3 | (A) Boxplots showing the wheat blast indices in 53 lines with the 2NS translocation in the diversity panel and 119 lines without the 2NS translocation in
the diversity panel. (B) Two-fold cross validation prediction accuracies for blast response in 53 lines with the 2NS translocation and 119 lines without the 2NS
translocation in the diversity panel using the fixed effects (Fixed), genomic best linear unbiased prediction (GBLUP), GBLUP and fixed effects (GBLUP + Fixed), Bayes
B, and pedigree best linear unbiased prediction (ABLUP) models. The prediction accuracies are missing for some environments and models in the subset of lines
with the 2NS translocation, where several lines had a blast index of zero. In (A,B), FP refers to the first planting, SP refers to the second planting and BLUEs refer to
the best linear unbiased estimates of blast indices across the different environments.
that the blast BLUEs had the highest mean prediction accuracy included markers 2A_718152 (718,152 bps, 0 cM), 2A_1686041
(0.56± 0.03) across all the models, and Okinawa 2018 SP had the (1,686,041 bps, 0 cM), 2A_1872142 (1,872,142 bps, 0 cM) and
lowest mean prediction accuracy (−0.04± 0.04). 2A_2367215 (2,367,215 bps, 0 cM), that were located between
718,152 and 2,367,215 bps on the Refseq v1.0 (IWGSC, 2018) and
Breeding Panel at 0 cM on the Popseq map (Chapman et al., 2015).
Statistical Analysis of Blast Indices in the Breeding The highest mean prediction accuracy with the different
Panel models in the breeding panel was observed in the blast BLUEs
Statistical analysis of blast indices in the breeding panel (Table 2) dataset (0.56-0.81). However, unlike in the diversity panel, the
indicated that the mean blast indices was the highest in accuracies in the blast BLUEs dataset from each model were only
Quirusillas 2019 FP (14.6 ± 27.5) and lowest in Quirusillas 2018 4.7–15.3% higher than the mean prediction accuracies of the
FP (10.2 ± 19.2). While the maximum blast indices ranged individual environments. Across all the models, we observed that
between 68.6 and 100 in the different datasets, 48% (Quirusillas the mean prediction accuracy was the highest in the Quirusillas
2019 SP) to 62.5% (Quirusillas 2018 FP) of the lines in the 2018 FP (0.66 ± 0.06) dataset and the lowest in Quirusillas 2019
different environments had a blast index of zero. The phenotypic FP (0.63± 0.11) dataset.
correlation between the blast indices in the Quirusillas 2019 FP The tests for the significance of the mean differences between
and SP was very high (0.82). The Okinawa 2018 SP dataset also the prediction accuracies obtained from the different models
had high correlations (ranged between 0.70 and 0.75) with the indicated that they were not significant in the two-tailed t-tests
Quirusillas blast evaluations. The narrow-sense heritability of at a threshold of 0.005 for the following model pairs:
blast across all the environments in the breeding panel was 0.65
σ2
A 318 and σ2
ε = 168).
TABLE 2 | Statistical analysis of blast indices in the breeding panel with 248 lines.
Prediction Accuracies for Blast in the Breeding Panel
Dataset Mean Standard Median Minimum Maximum
The mean prediction accuracies for blast in the breeding panel deviation
using different models were: (i) 0.75± 0.04 using the fixed effects
model (ii) 0.73 ± 0.05 using the GBLUP + fixed effects model Okinawa 2018 SP 10.6 18.1 0 68.6 0
(iii) 0.70 ± 0.02 using Bayes B model (iv) 0.61 ± 0.06 using Quirusillas 2018 FP 10.2 19.2 0 87.2 0
the GBLUP model and (v) 0.51 ± 0.06 using the ABLUP model Quirusillas 2019 FP 14.6 27.5 0 100 0
(Figure 4). In the fixed effects model, one to four selected markers Quirusillas 2019 SP 14.1 24.7 1 94.1 0
on chromosome 2AS (Supplementary Table 4) were used as BLUEs 12.4 20.1 2.3 77.9 0
fixed effects in the different datasets of the breeding panel. This FP, First planting; SP, Second planting; BLUEs, Best linear unbiased estimates.
Frontiers in Plant Science | www.frontiersin.org 8 January 2022 | Volume 12 | Article 745379
fpls-12-745379 January 3, 2022 Time: 12:58 # 9
Juliana et al. Genomic Selection for Wheat Blast
FIGURE 4 | Twofold cross validation prediction accuracies for blast response in the breeding panel (248 lines) using the fixed effects (Fixed), genomic best linear
unbiased prediction (GBLUP), GBLUP and fixed effects (GBLUP + Fixed), Bayes B, and pedigree best linear unbiased prediction (ABLUP) models. FP refers to the
first planting, SP refers to the second planting and BLUEs refer to the best linear unbiased estimates of blast indices across the different environments.
(i) GBLUP+ fixed effects and Bayes B: Mean difference = 0.03, the lines selected by PS and 57.1% lines overlapped with the lines
p-value = 0.13 discarded by PS.
(ii) GBLUP + fixed effects and fixed effects: Mean
difference = 0.02, p-value = 0.01 Blast Distribution and Prediction Accuracies in
(iii) Bayes B and fixed effects: Mean difference = 0.05, p- Subsets of Lines With and Without the 2NS
value = 0.02. Translocation in the Breeding Panel
In the 185 lines with the 2NS translocation in the breeding panel,
However, the Bayes B, GBLUP + fixed effects and fixed we observed that the mean blast index ranged between 2.3 ±
effects models had significantly higher prediction accuracies 5.4 and 3.7 ± 8.9 in the different environments (Figure 6A).
compared to the GBLUP model, with the mean differences In the 47 lines without the 2NS translocation in the breeding
ranging between 0.10 and 0.14 and the p-values ranging between panel, the mean blast index ranged between 38.2 ± 18.7 and
7.6 × 10−5 and 2.4 × 10−3. Similarly, all the marker-based 56.8 ± 30.3 in the different environments. The mean prediction
models had significantly higher prediction accuracies compared accuracies for blast across the different environments for the lines
to the ABLUP model, with the mean differences ranging between with the 2NS translocation in the breeding panel ranged between
0.10 and 0.25 and the p-values ranging between 3.6 × 10−6 and 0.27 ± 0.14 using the Bayes B model and 0.04 ± 0.19 using
5.3× 10−4. the GBLUP + fixed effects model (Figure 6B). Similarly, for the
lines without the 2NS translocation in the breeding panel, the
mean prediction accuracies ranged between 0.10 ± 0.04 using
Phenotypic Selection vs. Estimated Breeding Values the ABLUP model and 0.03± 0.08 using the Bayes B model. The
Based Selection for Blast in the Breeding Panel markers used as fixed effects in the different datasets for the lines
To compare PS and EBVs-based selection for blast resistance with and without the 2NS translocation in the breeding panel are
using the BLUEs dataset in the breeding panel, we selected 185 given in Supplementary Tables 5, 6, respectively.
lines (74.6%) with blast indices less than 10 and an equal number We observed that in the subsets of lines with and without
of lines using the EBVs (Figure 5). The highest percentage of the 2NS translocation, the mean prediction accuracy was not
overlap with PS was obtained using the EBVs from the fixed significantly different (mean difference = 0.05, p-value = 0.25). In
effects model, where 95.7% lines were selected by both MAS the breeding panel lines with the 2NS translocation, we observed
and PS, while 87.3% of the lines were not selected by both. that the blast BLUEs had the highest mean prediction accuracy
Selection from the GEBVs obtained from the GBLUP + fixed (0.32 ± 0.12) across all the models, and Quirusillas 2019 FP
effects, Bayes B and GBLUP models resulted in selection of 94.6, had the lowest mean prediction accuracy (−0.004 ± 0.15). In
94, and 90.3% lines, respectively, that were also selected by PS breeding panel lines without the 2NS translocation, the mean
and discarding of 84.1, 82.5, and 71.4% lines, respectively, that prediction accuracy across all the models was the highest in
were also discarded by PS. However, in pedigree selection using Quirusillas 2019 FP (0.13 ± 0.06) and the lowest in Quirusillas
the EBVs from the ABLUP model, 85.4% lines overlapped with 2019 SP (0.01± 0.07).
Frontiers in Plant Science | www.frontiersin.org 9 January 2022 | Volume 12 | Article 745379
fpls-12-745379 January 3, 2022 Time: 12:58 # 10
Juliana et al. Genomic Selection for Wheat Blast
FIGURE 5 | Comparison of phenotypic selection (PS) of the best linear unbiased estimates of blast indices across environments with: (i) marker assisted selection
(MAS) using the estimated breeding values (EBVs) obtained from the fixed effects model (fixed) (ii) genomic selection (GS) using the genomic estimated breeding
values (GEBVs) obtained from the genomic best-linear unbiased prediction (GBLUP) and Bayes B models (iii) GS + MAS using the GEBVs obtained from the GBLUP
and fixed effects (GBLUP + Fixed) model and (iv) pedigree selection (PedS) using the EBVs obtained from the pedigree best linear unbiased prediction (ABLUP)
model in the breeding panel comprising 248 lines.
FIGURE 6 | (A) Boxplots showing the wheat blast indices in 185 lines with the 2NS translocation in the breeding panel and 47 lines without the 2NS translocation in
the breeding panel. (B) Two-fold cross validation prediction accuracies for blast response in 185 lines with the 2NS translocation and 47 lines without the 2NS
translocation in the breeding panel using the fixed effects (Fixed), genomic best linear unbiased prediction (GBLUP), GBLUP and fixed effects (GBLUP + Fixed),
Bayes B, and pedigree best linear unbiased prediction (ABLUP) models. In (A,B), FP refers to the first planting, SP refers to the second planting and BLUEs refer to
the best linear unbiased estimates of blast indices across the different environments.
Caninde#1 × Alondra Full-Sibs Panel between the blast indices in Jashore and the Bolivian sites (ranged
Statistical Analysis of Blast Indices in the Caninde#1 between 0.39 and 0.69), while high to very high correlations
× Alondra Full-Sibs Panel (ranged between 0.58 and 0.82) were observed between the
In the Caninde#1 × Alondra full-sibs panel (Table 3), we blast indices in Okinawa and Quirusillas. The narrow-sense
observed that the mean blast indices were the highest in the heritability of blast across all the environments in the full-sibs
Okinawa 2019 FP (55.7± 41.8) dataset. While the maximum blast panel was 0.55 (σ2
A = 633.7 and σ2
ε = 520.9).
index was 100 in nine out of the 12 datasets, we also observed
that 15.8% (Jashore 2019 FP) to 42.3% (Quirusillas 2018 FP) Prediction Accuracies for Blast in the Caninde#1 ×
of the lines in the different datasets had a blast index of zero. Alondra Full-Sibs Panel
Across the different planting times, we observed moderate to high The mean prediction accuracies for blast in the Caninde#1 ×
correlations between the blast indices ranging between 0.87 in Alondra population using different models were: (i) 0.57 ±
Okinawa 2019 and 0.58 in Jashore 2018. Considering the different 0.10 using the fixed effects model (ii) 0.57 ± 0.10 using the
sites of blast evaluation, we observed moderate correlations GBLUP + fixed effects model (iii) 0.54 ± 0.10 using Bayes B
Frontiers in Plant Science | www.frontiersin.org 10 January 2022 | Volume 12 | Article 745379
fpls-12-745379 January 3, 2022 Time: 12:58 # 11
Juliana et al. Genomic Selection for Wheat Blast
TABLE 3 | Statistical analysis of blast indices in the Caninde#1 × Alondra full-sibs The two-tailed t-tests for the significance of the mean
panel with 298 lines. differences between the prediction accuracies obtained from
Dataset Mean Standard Median Minimum Maximum different models indicated that they were not significant at a
deviation threshold of 0.005 for the GBLUP + fixed effects and fixed
effects models (Mean difference = 0.005, p-value = 0.19). We also
Quirusillas 2018 FP 19.4 21.3 13.4 0 90.1 observed that the fixed effects, Bayes B and GBLUP+ fixed effects
Quirusillas 2018 SP 32.4 28.4 37.9 0 100 models had significantly higher prediction accuracies compared
Quirusillas 2019 FP 41.6 36.9 43.1 0 100 to the GBLUP model, with the mean differences ranging between
Quirusillas 2019 SP 43.6 38.4 48.1 0 100 0.05 and 0.08 and the p-values for the test of significance of the
Jashore 2018 FP 29.1 28.1 24.5 0 100 mean differences ranging between 3.2 × 10−7 and 2.9 × 10−8.
Jashore 2018 SP 28.5 24.8 24.7 0 100 Similarly, the prediction accuracies from the GBLUP + fixed
Jashore 2019 FP 32.3 25.3 30.8 0 100 effects and the fixed effects models were significantly higher than
Jashore 2019 SP 45.8 40.0 39.5 0 100 those from the Bayes B model with a mean difference of 0.03 and
Okinawa 2018 FP 34.8 28.6 42.6 0 92.9 the p-value for the test of significance of the mean differences
Okinawa 2018 SP 28.0 25.8 29.3 0 96.0 ranging between 1.9× 10−4 and 1.1× 10−5.
Okinawa 2019 FP 55.7 41.8 70.3 0 100
Okinawa 2019 SP 46.5 39.3 49.5 0 100 Phenotypic Selection vs. Estimated Breeding Value
BLUE 36.5 25.6 45.5 0 90.1 Based Selection for Blast in the Full-Sibs Panel
FP, First planting; SP, Second planting; BLUEs, Best linear unbiased estimates. The blast BLUEs dataset was used to select 82 lines (27.5%) with
BLUEs less than 10 and a similar number of lines were selected
from the EBVs obtained from different models (Figure 8). We
model and (iv) 0.49 ± 0.10 using the GBLUP model (Figure 7). observed that MAS based on EBVs from the fixed effects model
In the fixed effects model, one to three selected markers on had the highest percentage of overlap with PS, resulting in 75.6%
chromosome 2AS (Supplementary Table 7) were used as fixed of the lines selected by both and 90.7% of the lines discarded by
effects and they included the STS markers (cslVrgal3, IWB11136, both methods. The GEBVs obtained from the GBLUP + fixed
Ventriup, WGGB156, and WGGB159) and the GBS marker, effects, Bayes B and GBLUP models resulted in selection of 59.8,
2A_14418709 (14,418,709 bps and 8.9 cM). Similar to the 57.3, and 58.5% lines, respectively, that were also selected by PS
diversity and breeding panels, the highest mean prediction and discarding of 84.7, 83.8, and 84.3% lines, respectively, that
accuracies with the different models in the full-sibs panel was were also discarded by PS.
observed in the blast BLUEs dataset (0.65–0.72), that were
29.2–36.8% higher than the mean prediction accuracies of the Blast Distribution and Prediction Accuracies in
individual environments. When the mean prediction accuracies Subsets of Lines With and Without the 2NS
of the environments across all the models were considered, we Translocation in the Full-Sibs Panel
observed that it was the highest in Okinawa 2019 FP (0.68± 0.04) In the 117 full-sibs with the 2NS translocation, we observed that
dataset and lowest in Jashore 2019 FP (0.41± 0.03) dataset. the mean blast index ranged between 7.8 ± 16.1 and 19.5 ± 31.3
FIGURE 7 | Twofold cross validation prediction accuracies for blast response in the full-sibs panel (298 lines) using the fixed effects (Fixed), genomic best linear
unbiased prediction (GBLUP), GBLUP and fixed effects (GBLUP + Fixed) and Bayes B models. FP refers to the first planting, SP refers to the second planting and
BLUEs refer to the best linear unbiased estimates of blast indices across the different environments.
Frontiers in Plant Science | www.frontiersin.org 11 January 2022 | Volume 12 | Article 745379
fpls-12-745379 January 3, 2022 Time: 12:58 # 12
Juliana et al. Genomic Selection for Wheat Blast
in the different environments (Figure 9A). In the 144 full-sibs p-value = 5.7 × 10−4). In the full-sibs panel lines with the
without the 2NS translocation, the mean blast index ranged 2NS translocation, we observed that Quirusillas 2018 FP had the
between 29± 20.8 and 82.9± 26.2 in the different environments. highest mean prediction accuracy (0.17 ± 0.05) across all the
The mean prediction accuracies for blast across the different models, and Okinawa 2018 FP had the lowest mean prediction
environments for the lines with the 2NS translocation in the full- accuracy (−0.14 ± 0.08). In full-sib panel lines without the 2NS
sibs panel ranged between 0.03 ± 0.09 using the Bayes B model translocation, the mean prediction accuracy across all the models
and−0.02± 0.11 using the GBLUP model (Figure 9B). Similarly, was the highest in Jashore 2019 FP (0.24 ± 0.07) and the lowest
for the lines without the 2NS translocation in the full-sibs panel, in Jashore 2019 SP (−0.05± 0.03).
the mean prediction accuracies ranged between 0.15± 0.12 using
the GBLUP model and 0.04 ± 0.08 using the fixed effects model.
The markers used as fixed effects in the different datasets for the DISCUSSION
lines with and without the 2NS translocation in the full-sibs panel
are given in Supplementary Tables 8, 9, respectively. We have successfully evaluated genomic prediction for wheat
We observed that the mean prediction accuracy across all the blast in three panels using the GBLUP and Bayes B models and
environments and models was significantly higher in the subset compared the genomic prediction accuracies with those from
of lines without the 2NS translocation compared to the subset the fixed effects, GBLUP + fixed effects and ABLUP models,
of lines with the 2NS translocation (mean difference = 0.07, to understand the relative advantage of using genome-wide
FIGURE 8 | Comparison of phenotypic selection (PS) of the best linear unbiased estimates of blast indices across environments with: (i) marker assisted selection
(MAS) using the estimated breeding values (EBVs) obtained from the fixed effects model (fixed) (ii) genomic selection (GS) using the genomic estimated breeding
values (GEBVs) obtained from the genomic best-linear unbiased prediction (GBLUP) and Bayes B models and (iii) GS + MAS using the GEBVs obtained from the
GBLUP and fixed effects (GBLUP + Fixed) model in the Caninde#1 × Alondra full-sibs panel comprising 298 lines.
FIGURE 9 | (A) Boxplots showing the wheat blast indices in 117 lines with the 2NS translocation in the full-sibs panel and 144 lines without the 2NS translocation in
the full-sibs panel. (B) Twofold cross validation prediction accuracies for blast response in 117 lines with the 2NS translocation and 144 lines without the 2NS
translocation in the full-sibs panel using the fixed effects (Fixed), genomic best linear unbiased prediction (GBLUP), GBLUP and fixed effects (GBLUP + Fixed) and
Bayes B models. In (A,B), FP refers to the first planting, SP refers to the second planting and BLUEs refer to the best linear unbiased estimates of blast indices
across the different environments.
Frontiers in Plant Science | www.frontiersin.org 12 January 2022 | Volume 12 | Article 745379
fpls-12-745379 January 3, 2022 Time: 12:58 # 13
Juliana et al. Genomic Selection for Wheat Blast
markers. On average, across all the panels and environments selection pressure on the MoT populations (Cruz and Valent,
analyzed in this study, the GBLUP + fixed effects model (0.63 ± 2017; Cruppe et al., 2020). However, in such cases where there
0.13) and the fixed effects model (0.62 ± 0.13) were the best is a risk of resistance breakdown and narrowing down the genetic
models for predicting blast, followed by the Bayes B (0.59± 0.11), variation for blast resistance by using predictions based on only
GBLUP (0.55 ± 0.1), and ABLUP (0.48 ± 0.06) models. Our one locus, the 2NS translocation-based markers can still be used
results also indicated that there was no significant difference in to predict and select against the translocation.
the prediction accuracies from the GBLUP and Bayes B genomic Comparison of genomic and pedigree-based prediction
prediction models in the diversity panel, as also observed in accuracies indicated that in both the diversity and breeding
previous studies (Heslot et al., 2012; Juliana et al., 2017b). panels, the ABLUP model resulted in the lowest prediction
However, in the other two panels, the Bayes B model gave accuracies. This is consistent with previous studies that have
significantly higher accuracies compared to the GBLUP model, reported superiority of genomic prediction over prediction
probably because the Bayes B model assumptions fitted well the prediction (Crossa et al., 2010; Spindel et al., 2015), while other
genetic architecture of blast response in these panels, where the studies have also reported similar accuracies from both (Juliana
2NS translocation had a large effect. et al., 2017a,b, 2018, 2020b). However, we also observed that the
On comparing blast prediction accuracies from the genomic ABLUP blast prediction accuracies were 85.4 and 83.6% of the
prediction models (GBLUP and Bayes B) with prediction mean genomic prediction accuracies from the Bayes B model
accuracies from the fixed effects model and the GBLUP + fixed in the different datasets of the diversity panel and breeding
effects model, we observed: (i) no significant differences between panel, respectively. This implies that although pedigree-based
the fixed effects, GBLUP + fixed effects and the Bayes B prediction for blast does not result in the highest accuracy,
models in both the diversity and breeding panels and (ii) pedigree relationships can also be useful in predicting blast
significantly higher prediction accuracies from the fixed effects resistance, when genotyping data is not available or affordable.
model and GBLUP + fixed effects model compared to the Among the three sites of blast evaluation and prediction
genomic prediction models in the full-sibs panel. These results in this study, our results showed that Okinawa (0.63 ± 0.09)
are contrasting to previous studies that have reported the had the highest mean prediction accuracy across the different
superiority of genomic prediction models over the fixed effects panels, models, years and planting times, followed by Quirusillas
model for some diseases in wheat (Rutkoski et al., 2012; Juliana (0.59 ± 0.1) and Jashore (0.45 ± 0.06). Our results also indicated
et al., 2017b) and higher accuracies by integrating genomic that the blast BLUEs dataset was the best predicted in all the
prediction and the fixed effects model (Odilbekov et al., 2019; three panels and accuracies in the BLUEs datasets were 4.7–
Sehgal et al., 2020). However, given that blast response in all the 45.5% higher than the mean prediction accuracies observed
panels in this study was predominantly controlled by the 2NS in the individual environments. One possible explanation to
translocation (He et al., 2020, 2021; Juliana et al., 2020a), our this is that the BLUEs obtained from multi-environment
results are in agreement with Juliana et al. (2017a), who reported evaluations are most likely to be close to the true breeding
that for seedling leaf and stripe rust resistance, where a single gene values of the genotypes and hence predicted with the highest
had a large effect on the disease response in the population, the accuracy, thereby making them more robust for utilization in
fixed effects model and the GBLUP+ fixed effects model perform predictive breeding, compared to single-environment phenotypic
similar to or slightly better than the genomic prediction models. observations. Another interesting observation in our study
Hence, our findings have important implications for wheat blast was that across all the environments, panels and models, the
predictions in environments where the resistance is determined prediction accuracies from the two planting times were not
by the 2NS translocation and indicate that in such environments, significantly different (FP mean prediction accuracy: 0.56± 0.12;
a fixed effects model with one to few markers tagging the 2NS SP mean prediction accuracy: 0.57± 0.11), indicating that highly
translocation would be sufficient and genome-wide markers may correlated blast indices in different planting times result in similar
not lead to a significant increase in blast prediction accuracies. prediction accuracies.
The 2NS translocation linked markers that were effective Among the three panels evaluated for wheat blast prediction,
in predicting blast response in more than a fold or dataset the breeding panel had the highest mean prediction accuracy
in this study included the Illumina Infinium 15K BeadChip (0.66 ± 0.1), followed by the diversity panel (0.59 ± 0.13) and
markers, Kukri_c22599_114 and Tdurum_contig29983_490 (El full-sibs panel (0.54 ± 0.1). This is a promising outcome of this
Hanafi et al., 2021); GBS markers, 2A_1686041, 2A_1872142, study indicating that blast can be predicted with moderate to
2A_718152 and 2A_14418709 and STS markers, cslVrgal3, high predictabilities in all these panels, and hence prediction-
IWB11136, Ventriup, WGGB156, and WGGB159, all of which based selection for wheat blast can be successfully implemented
can be used to select for the 2NS translocation based blast in any historic germplasm, breeding lines and sister lines.
resistance. But, it should also be noted that the 2NS translocation- However, we could not directly compare prediction accuracies
based blast resistance is incomplete and sometimes background- across panels, because of the different sizes of the panels, the
dependent (Cruz et al., 2016b; Cruppe, 2020), with reports of different genotyping platforms used and also the different blast
the MoT isolates in Brazil (Ceresini et al., 2018) and Paraguay distributions in these panels. For example, the breeding panel
(Singh et al., 2016) having overcome the 2NS translocation-based had the highest number of resistant lines (48–62.5%) with a
blast resistance and hence relying on only one large effect blast index of zero and this might have also contributed to high
resistance locus is not recommended, as it could result in prediction accuracies.
Frontiers in Plant Science | www.frontiersin.org 13 January 2022 | Volume 12 | Article 745379
fpls-12-745379 January 3, 2022 Time: 12:58 # 14
Juliana et al. Genomic Selection for Wheat Blast
This study is also unique because three different whole- of lines were significantly lower than the mean prediction
genome marker platforms, the Illumina Infinium 15K BeadChip, accuracies obtained in the full set of lines in each panel, our
GBS and DArTseq were evaluated for predicting wheat blast. results demonstrate the possibility of implementing GS for blast
Considering only the two genomic prediction models (GBLUP in panels of lines without the 2NS translocation. However, it
and Bayes B), we observed that the breeding panel genotyped should be noted that our observations of blast predictions in lines
using GBS was the best predicted (0.66 ± 0.07), followed by with and without the 2NS translocation were done using subsets
the diversity panel genotyped using the Illumina Infinium 15K of few lines (53, 185, and 117 lines from the three panels had the
BeadChip (0.60 ± 0.09) and the full-sibs panel genotyped using 2NS translocation and 119, 47, and 144 lines from the three panels
the DArTseq platform (0.51 ± 0.1). While previous studies that did not have the 2NS translocation), and hence larger panels
have reported the superiority of GBS over both the DArTseq are needed to further understand the prediction accuracies for
(Juliana et al., 2017b) and array-based platforms (Elbasyoni blast in panels of lines with and without the 2NS translocation.
et al., 2018), the differences in prediction accuracies using The higher blast predictabilities in the subsets of lines without the
these three platforms cannot be compared per se in this study, 2NS translocation could be because of the low variability in the
because of the aforementioned reasons (different panel sizes blast indices in the lines with the 2NS translocation (mean blast
and blast distributions across panels) and none of the panels indices ranged between 1.5 and 19.5) and the moderate to high
were genotyped using all the platforms. Hence, further studies variability in the blast indices (mean blast indices ranged between
on genomic predictions in different panels genotyped using 14.3 and 82.9) in subsets without the 2NS translocation. We also
the same genotyping platform are essential to compare blast observed that the mean prediction accuracies using the fixed
predictabilities across different panels. Using a common platform effects model were very low (less than 0.10 in most subsets except
to genotype different panels would also be useful to explore in the lines without the 2NS translocation in the diversity panel),
beyond the cross-validation strategy evaluated in this study and and the markers that were used in the different folds and datasets
evaluate genomic prediction for blast across panels to understand of the fixed effects model were inconsistent, indicating that the
how well one panel can be predicted from another. This would be fixed effects model is not an ideal choice when there are no large
akin to a practical GS implementation scenario, where breeders effect consistent markers associated with blast in the panels.
would be interested in predicting the blast response of lines from Overall, this study has provided important insights into
new panels using any existing panel. Since genomic prediction the genomic predictability of wheat blast and the prospects
accuracies be lower in across-panel predictions compared to of implementing GS and MAS for the disease. One caveat
within-panel predictions (Juliana et al., 2019), it is important to in this study is that in all the three panels, blast resistance
evaluate across-panel predictions for wheat blast. was controlled to a large extent by the 2NS translocation and
This study was also aimed to test the hypothesis that GS would hence further studies on genomic prediction of quantitative blast
perform better than MAS and pedigree-based selection for wheat resistance in panels where resistance is not controlled by the 2NS
blast. On average, across all the datasets and panels, MAS led to translocation is needed. In conclusion, we have demonstrated
the selection of the highest percentage (88.5%) of lines selected that in populations where blast resistance is controlled by the
by PS and discard of the highest percentage of lines (91.8%) that 2NS translocation, MAS using few markers tagging the 2NS
were discarded by PS. In contrast, on average, GS GBLUP and GS translocation can be used for accelerating predictive breeding
Bayes B only led to the selection of 75.2 and 80.6% of the lines that for blast. This is a key finding of this study that opens several
were selected by PS and discard of 81.9 and 87.4% of the lines that opportunities for wheat breeding programs to:
were discarded by PS, respectively. These results clearly indicated
that MAS outperformed GS in our study, despite the phenotypic (i) Screen a subset of lines in the blast hot-spots and use that
responses being continuous and indicating quantitative genetic phenotyping data to predict the blast breeding values for
control. However, pedigree-based selection, on average led to the other related lines, as demonstrated in this study where we
selection of 77.3% of the lines that were selected by PS and the evaluated genomic prediction assuming that a half of the
discard of 71.9% of lines that were discarded by PS and hence lines were phenotyped.
GS was superior to pedigree-based selection as hypothesized. It is (ii) Use the predicted breeding values to complement
also interesting that in a previous study comparing GS and PS for selection based on the phenotype and increase the
grain yield which is a highly quantitative trait, GS could select a selection accuracy.
maximum of 70.9% of the top lines and discard 71.5% of the poor (iii) Use the 2NS translocation-associated molecular markers
lines (Juliana et al., 2018) at a selection intensity of 0.5, which to select for or against the 2NS based-blast resistance
is significantly lower than the percentage overlap with PS in this without phenotyping.
study, owing to less complex genetic architecture of wheat blast (iv) Scale-up selection for blast resistance to early generations
resistance in the panels used in this study. of the breeding program that have been genotyped, but
We compared the prediction accuracies from different models are in large numbers to be phenotyped. For example,
obtained from subsets of lines with and without the 2NS the CIMMYT global wheat program screens international
translocation and the mean prediction accuracies across the nurseries (200-300 lines) derived from the stage 3 yield
different panels were 0.03 ± 0.16 (ranged between −0.22 and trials for blast resistance, but about 9,000 stage 1 yield
0.45) and 0.16 ± 0.18 (ranged between −0.09 and 0.57), trial lines are genotyped each year. Here, the international
respectively. While the mean prediction accuracies in the subsets nurseries can be used as training populations to predict the
Frontiers in Plant Science | www.frontiersin.org 14 January 2022 | Volume 12 | Article 745379
fpls-12-745379 January 3, 2022 Time: 12:58 # 15
Juliana et al. Genomic Selection for Wheat Blast
blast breeding values of the large set of stage 1 yield trial AUTHOR CONTRIBUTIONS
lines, thereby saving substantial cost and resources. In this
case, GS can provide an advantage over MAS, as the same PJ performed the analyses and drafted the first draft of the
genotyping data can be used to select for multiple traits in manuscript. PS and RS designed the study and supervised
the early generations. the analysis. XH, RI, BA, and FM were involved in blast
(v) Sparse-test genotypes in different blast hot-spots in which phenotyping. JP and SS were involved in generating the
not all the genotypes are grown in all the environments genotyping data. GS, AC, PS, RS, and AJ provided germplasm
(Jarquin et al., 2020). For example: when there are cost- and funds. All authors reviewed and approved the final version
constraints for breeding programs to evaluate blast in of the manuscript.
multiple sites, then sparse-testing can be implemented in
correlated sites.
(vi) For non-2NS resistance based predictive breeding, since
screening a large number of lines for blast in field FUNDING
conditions to build training sets is challenging, greenhouse
testing of blast can be used to primarily identify new This research was supported by the Accelerating Genetic Gain
resistance genes. This can be followed by obtaining (AGG) in Mazie and Wheat Project Grant INV-003439 [funded
GEBVs of the selected lines and then the best lines using by the Bill and Melinda Gates Foundation and the Foreign
PS and GS can be advanced for multilocation testing. and Commonwealth Development Office (FCDO)], Feed the
Simultaneous selection against the 2NS translocation can Future project # AID-OAA-A-13-00051 through the U.S. Agency
also be performed using molecular markers, to facilitate the for International Development (USAID), USAID-CtEH-AGG
identification of non-2NS based resistance. Supplement grant, CGIAR Research Program on WHEAT, Indian
Council of Agricultural Research, India, Vetenskapsrådet (the
Swedish Research Council), Sweden, and Australian Centre for
CONCLUSION International Agricultural Research, Australia #CIM/2016/219.
In conclusion, while this study demonstrates the potential of
MAS and GS for wheat blast resistance breeding, we would
also like to emphasize that continued efforts to use genomic
tools to identify non-2NS based sources of blast resistance in SUPPLEMENTARY MATERIAL
wheat is critical, which will involve coordinated high-throughput The Supplementary Material for this article can be found
genomics and phenomics approaches. online at: https://www.frontiersin.org/articles/10.3389/fpls.2021.
745379/full#supplementary-material
DATA AVAILABILITY STATEMENT
Supplementary Data 1 | Wheat blast indices of the lines in the diversity, breeding
The original contributions presented in the study are included and full-sibs panel that were evaluated in Jashore, Okinawa and Quirusillas in two
different planting dates indicated as first planting (FP) and second planting (SP)
in the article/Supplementary Material, further inquiries can be and the datasets are referred to by the site followed by the harvest year and
directed to the corresponding author/s. planting time.
REFERENCES Cardoso, C. A. D. A., Reis, E. M., and Moreira, E. N. (2008). Development
of a warning system for wheat blast caused by Pyricularia grisea. Summa
Anh, V. L., Anh, N. T., Tagle, A. G., Vy, T. T. P., Inoue, Y., Takumi, S., et al. (2015). Phytopathol. 34, 216–221. doi: 10.1590/S0100-54052008000300002
Rmg8, a new gene for resistance to Triticum isolates of Pyricularia oryzae in Castroagudín, V. L., Ceresini, P. C., De Oliveira, S. C., Reges, J. T. A., Maciel,
hexaploid wheat. Phytopathology 105, 1568–1572. doi: 10.1094/PHYTO-02-15- J. L. N., Bonato, A. L. V., et al. (2015). Resistance to QoI fungicides is
0034-R widespread in Brazilian populations of the wheat blast pathogen Magnaporthe
Anh, V. L., Inoue, Y., Asuke, S., Vy, T. T. P., Anh, N. T., Wang, S., et al. (2018). oryzae. Phytopathology 105, 284–294. doi: 10.1094/PHYTO-06-14-
Rmg8 and Rmg7, wheat genes for resistance to the wheat blast fungus, recognize 0184-R
the same avirulence gene AVR-Rmg8. Mol. Plant Pathol. 19, 1252–1256. doi: Ceresini, P. C., Castroagudín, V. L., Rodrigues, F. Á, Rios, J. A., Eduardo Aucique-
10.1111/mpp.12609 Pérez, C., Moreira, S. I, et al. (2018). Wheat blast: past, present, and future.
Barea, G., and Toledo, J. (1996). Identificación y Zonificación de Pyricularia o Annu. Rev. Phytopathol. 56, 427–456. doi: 10.1146/annurev-phyto-080417-
Brusone (Pyricularia oryzae) en el Cutivo De Trigo en el Departamento de 050036
Santa Cruz. Santa Cruz de la Sierra: Centro de Investigación Agrícola Tropical, Chapman, J. A., Mascher, M., Buluç, A., Barry, K., Georganas, E., Session, A., et al.
76–86. (2015). A whole-genome shotgun approach for assembling and anchoring the
Bradbury, P. J., Zhang, Z., Kroon, D. E., Casstevens, T. M., Ramdoss, Y., hexaploid bread wheat genome. Genome Biol. 16:26. doi: 10.1186/s13059-015-
and Buckler, E. S. (2007). TASSEL: Software for association mapping of 0582-8
complex traits in diverse samples. Bioinformatics 23, 2633–2635. doi: 10.1093/ Chowdhury, A. K., Saharan, M. S., Aggrawal, R., Malaker, P. K., Barma, N. C. D.,
bioinformatics/btm308 Tiwari, T. P., et al. (2017). Occurrence of wheat blast in Bangladesh and its
Cabrera, M., and Gutiérrez, S. (2007). Primer Registro de Pyricularia Grisea en implications for South Asian wheat production. Indian J. Genet. Plant Breed.
Cultivos de Trigo del NE de Argentina. Buenos Aires: IFSC Press. 77, 1–9. doi: 10.5958/0975-6906.2017.00001.3
Frontiers in Plant Science | www.frontiersin.org 15 January 2022 | Volume 12 | Article 745379
fpls-12-745379 January 3, 2022 Time: 12:58 # 16
Juliana et al. Genomic Selection for Wheat Blast
Coelho, M. A., de, O., Torres, G. A. M., Cecon, P. R., and Santana, F. M. (2016). sequencing analysis pipeline. PLoS One 9:e90346. doi: 10.1371/journal.pone.009
Sowing date reduces the incidence of wheat blast disease. Pesqui. Agropecuária 0346
Bras. 51, 631–637. doi: 10.1590/s0100-204x2016000500025 Goddard, R., Steed, A., Chinoy, C., Ferreira, J. R., Scheeren, P. L., MacIel, J. L. N.,
Couch, B. C., and Kohn, L. M. (2002). A multilocus gene genealogy concordant et al. (2020). Dissecting the genetic basis of wheat blast resistance in the
with host preference indicates segregation of a new species, Magnaporthe Brazilian wheat cultivar BR 18-Terena. BMC Plant Biol. 20:398. doi: 10.1186/
oryzae, from M. grisea. Mycologia 94, 683–693. doi: 10.1080/15572536.2003. s12870-020-02592-0
11833196 Goulart, A. C. P., Paiva, F., de, A., and de Mesquita, A. N. (1990). Occurrence
Crossa, J., de los Campos, G., Pérez, P., Gianola, D., Burgueño, J., Araus, J. L., et al. of Wheat Blast (Pyricularia Oyzae) in the State Of Mato Grosso do Sul, Brazil.
(2010). Prediction of genetic values of quantitative traits in plant breeding using Manhattan: Kansas State University.
pedigree and molecular markers. Genetics 186, 713–724. doi: 10.1534/genetics. Goulart, A. C. P., Sousa, P. G., and Urashima, A. S. (2007). Danos em trigo
110.118521 causados pela infecção de Pyricularia grisea. Summa Phytopathol. 33, 358–363.
Cruppe, G. (2020). Wheat Blast Management Through Identification of Novel doi: 10.1590/s0100-54052007000400007
Sources of Genetic Resistance and Understanding of Disease Dynamics. Habier, D., Fernando, R. L., and Garrick, D. J. (2013). Genomic BLUP decoded:
Dissertation. Manhattan: Kansas State University. a look into the black box of genomic prediction. Genetics 194, 597–607. doi:
Cruppe, G., Cruz, C. D., Peterson, G., Pedley, K., Asif, M., Fritz, A., et al. 10.1534/genetics.113.152207
(2020). Novel sources of wheat head blast resistance in modern breeding lines He, X., Juliana, P., Kabir, M. R., Roy, K. K., Islam, R., Marza, F., et al. (2021).
and wheat wild relatives. Plant Dis. 104, 35–43. doi: 10.1094/PDIS-05-19- Screening and mapping for head blast resistance in a panel of CIMMYT and
0985-RE South Asian bread wheat germplasm. Front. Genet. 12:679162. doi: 10.3389/
Cruz, C. D., Kiyuna, J., Bockus, W. W., Todd, T. C., Stack, J. P., and Valent, fgene.2021.679162
B. (2015). Magnaporthe oryzae conidia on basal wheat leaves as a potential He, X., Rezaul, M., Krishna, K., Babul, K. R., Kaijie, A., Felix, X., et al. (2020).
source of wheat blast inoculum. Plant Pathol. 64, 1491–1498. doi: 10.1111/ppa. QTL mapping for field resistance to wheat blast in the Caninde # 1 / Alondra
12414 population. Theor. Appl. Genet. 133, 2673–2683. doi: 10.1007/s00122-020-
Cruz, C. D., Magarey, R. D., Christie, D. N., Fowler, G. A., Fernandes, J. M., Bockus, 03624-x
W. W., et al. (2016a). Climate suitability for Magnaporthe oryzae Triticum Heffner, E. L., Lorenz, A. J., Jannink, J. L., and Sorrells, M. E. (2010). Plant breeding
pathotype in the United States. Plant Dis. 100, 1979–1987. doi: 10.1094/PDIS- with genomic selection: Gain per unit time and cost. Crop Sci. 50, 1681–1690.
09-15-1006-RE doi: 10.2135/cropsci2009.11.0662
Cruz, C. D., Peterson, G. L., Bockus, W. W., Kankanala, P., Dubcovsky, J., Jordan, Heffner, E. L., Sorrells, M. E., and Jannink, J.-L. (2009). Genomic selection for crop
K. W., et al. (2016b). The 2NS translocation from Aegilops ventricosa confers improvement. Crop Sci. 49, 1–12. doi: 10.2135/cropsci2008.08.0512
resistance to the Triticum pathotype of Magnaporthe oryzae. Crop Sci. 56, Helguera, M., Khan, I. A., Kolmer, J., Lijavetzky, D., Dubcovsky, J., and Jua, M.
990–1000. doi: 10.2135/cropsci2015.07.0410 (2003). PCR assays for the Lr37-Yr17-Sr38 cluster of rust resistance genes and
Cruz, C. D., Santana, F. M., Todd, T. C., Maciel, J. L. N., Kiyuna, J., Baldelomar, their use. Crop Sci 43, 1839–1847. doi: 10.1007/978-3-319-18263-6_8
D. F., et al. (2019). Multi-environment assessment of fungicide performance for Heslot, N., Yang, H.-P., Sorrells, M. E., and Jannink, J.-L. (2012). Genomic selection
managing wheat head blast (WHB) in Brazil and Bolivia. Trop. Plant Pathol. 44, in plant breeding: a comparison of models. Crop Sci. 52, 146–160. doi: 10.2135/
183–191. doi: 10.1007/s40858-018-0262-9 cropsci2011.09.0297
Cruz, C. D., and Valent, B. (2017). Wheat blast disease: danger on the Hossain, A., Mottaleb, K. A., Farhad, M., and Deb Barma, N. C. (2019). Mitigating
move. Trop. Plant Pathol. 42, 210–222. doi: 10.1007/s40858-017- the twin problems of malnutrition and wheat blast by one wheat variety, “BARI
0159-z Gom 33”, in Bangladesh. Acta Agrobot. 72:1775. doi: 10.5586/aa.1775
de los Campos, G., Vazquez, A. I, Fernando, R., Klimentidis, Y. C., and Sorensen, Igarashi, S. (1986). Pyricularia em trigo. 1. Ocorrencia de Pyricularia sp noestado
D. (2013). Prediction of complex human traits using the genomic best do Parana. Fitopatol. Bras. 11, 351–352.
linear unbiased predictor. PLoS Genet. 9:e1003608. doi: 10.1371/journal.pgen. Igarashi, S. (1990). “Update on wheat blast (Pyricularia oryzae) in Brazil,” in A
1003608 Proceeding of the International Conference-Wheat for the Nontaditional Warm
Dos Anjos, J. R. N., Da Silva, D. B., Charchar, M. J. D., and Rodrigues, G. C. (1996). Areas (Mexico: CIMMYT), 480–483.
Occurrence of blast fungus (Pyricularia grisea) on wheat and rye in the Savanna Islam, M. T., Croll, D., Gladieux, P., Soanes, D. M., Persoons, A., Bhattacharjee,
region on Central Brazil. Pesqui. Agropecuária Bras. 31, 79–82. P., et al. (2016). Emergence of wheat blast in Bangladesh was caused by a
El Hanafi, S., Backhaus, A., Bendaou, N., Sanchez-Garcia, M., Al-Abdallat, A., and South American lineage of Magnaporthe oryzae. BMC Biol. 14:84. doi: 10.1186/
Tadesse, W. (2021). Genome-wide association study for adult plant resistance to s12915-016-0309-7
yellow rust in spring bread wheat (Triticum aestivum L.). Euphytica 217, 1–14. Islam, M. T., Kim, K. H., and Choi, J. (2019). Wheat blast in Bangladesh: The
doi: 10.1007/s10681-021-02803-1 current situation and future impacts. Plant Pathol. J. 35, 1–10. doi: 10.5423/
Elbasyoni, I. S., Lorenz, A. J., Guttieri, M., Frels, K., Baenziger, P. S., Poland, J., PPJ.RW.08.2018.0168
et al. (2018). A comparison between genotyping-by-sequencing and array-based IWGSC (2018). Shifting the limits in wheat research and breeding using a fully
scoring of SNPs for genomic prediction accuracy in winter wheat. Plant Sci. 270, annotated reference genome. Science 361:661. doi: 10.1126/science.aar7191
123–130. doi: 10.1016/j.plantsci.2018.02.019 Jarquin, D., Howard, R., Crossa, J., Beyene, Y., Gowda, M., Martini, J. W. R., et al.
Endelman, J. B. (2011). Ridge regression and other kernels for genomic (2020). Genomic prediction enhanced sparse testing for multi-environment
selection with R Package rrBLUP. Plant Genome 4, 250–255. doi: 10.3835/ trials. G3 Genes Genomes Genet. 10, 2725–2739. doi: 10.1534/g3.120.40
plantgenome2011.08.0024 1349
Ferreira, J. R., Torres, G. A. M., Consoli, L., Camilotti, G. A., Scagliusi, S. M. M., Juliana, P., He, X., Kabir, M. R., Roy, K. K., Anwar, M. B., Marza, F., et al. (2020a).
Nhani, A., et al. (2021). Quantitative trait loci conferring blast resistance in Genome-wide association mapping for wheat blast resistance in CIMMYT’s
hexaploid wheat at adult plant stage. Plant Pathol. 70, 100–109. doi: 10.1111/ international screening nurseries evaluated in Bolivia and Bangladesh. Sci. Rep.
ppa.13278 10, 1–14. doi: 10.1038/s41598-020-72735-8
Gilmour, A. R. (1997). ASREML for testing fixed effects and estimating multiple Juliana, P., Singh, R. P., Braun, H. J., Huerta-Espino, J., Crespo-Herrera, L.,
trait variance components. Proc. Assoc. Adv. Anim. Breed. Genet. 12, 386–390. Govindan, V., et al. (2020b). Genomic selection for grain yield in the CIMMYT
Gilmour, A. R., Thompson, R., and Cullis, B. R. (1995). Average information wheat breeding program—Status and perspectives. Front. Plant Sci. 11:564183.
REML: An efficient algorithm for variance parameter estimation in doi: 10.3389/fpls.2020.564183
linear mixed models. Biometrics 51, 1440–1450. doi: 10.2307/253 Juliana, P., Poland, J., Huerta-Espino, J., Shrestha, S., Crossa, J., Crespo-Herrera, L.,
3274 et al. (2019). Improving grain yield, stress resilience and quality of bread wheat
Glaubitz, J. C., Casstevens, T. M., Lu, F., Harriman, J., Elshire, R. J., using large-scale genomics. Nat. Genet. 51, 1530–1539. doi: 10.1038/s41588-
Sun, Q., et al. (2014). TASSEL-GBS?: A high capacity genotyping by 019-0496-6
Frontiers in Plant Science | www.frontiersin.org 16 January 2022 | Volume 12 | Article 745379
fpls-12-745379 January 3, 2022 Time: 12:58 # 17
Juliana et al. Genomic Selection for Wheat Blast
Juliana, P., Singh, R. P., Poland, J., Mondal, S., Crossa, J., Montesinos-López, O. A., botanical and bio-agent) in controlling wheat blast in a blast prone area in
et al. (2018). Prospects and challenges of applied genomic selection—A new Bangladesh. Arch. Phytopathol. Plant Prot. 54, 252–260. doi: 10.1080/03235408.
paradigm in breeding for grain yield in bread wheat. Plant Genome 11, 1–17. 2020.1827652
doi: 10.1136/bmj.2.1403.1129-a Rutkoski, J., Benson, J., Jia, Y., Brown-Guedira, G., Jannink, J.-L., and Sorrells, M.
Juliana, P., Singh, R. P., Singh, P. K., Crossa, J., Rutkoski, J. E., Poland, J. A., et al. (2012). Evaluation of genomic prediction methods for Fusarium head blight
(2017b). Comparison of models and whole-genome profiling approaches for resistance in wheat. Plant Genome J. 5:51. doi: 10.3835/plantgenome2012.02.
genomic-enabled prediction of Septoria tritici blotch, Stagonospora nodorum 0001
blotch, and tan spot resistance in wheat. Plant Genome 10:82. doi: 10.3835/ Rutkoski, J. E., Poland, J. A., Singh, R. P., Huerta-espino, J., Bhavani, S., Barbier,
plantgenome2016.08.0082 H., et al. (2014). Genomic selection for quantitative adult plant stem rust
Juliana, P., Singh, R. P., Singh, P. K., Crossa, J., Huerta-Espino, J., Lan, C., et al. resistance in wheat. Plant Genome J. 6, 1–44. doi: 10.3835/plantgenome2014.02.
(2017a). Genomic and pedigree-based prediction for leaf, stem, and stripe rust 0006
resistance in wheat. Theor. Appl. Genet. 130, 1415–1430. doi: 10.1007/s00122- Sadat, M. A., and Choi, J. (2017). Wheat blast: A new fungal inhabitant to
017-2897-1 Bangladesh threatening world wheat production. Plant Pathol. J. 33, 103–108.
Kohli, M. M., Mehta, Y. R., Guzman, E., de Viedma, L., and Cubilla, L. E. (2011). doi: 10.5423/PPJ.RW.09.2016.0179
Pyricularia blast-a threat to wheat cultivation. Czech J. Genet. Plant Breed. 47, Seah, S., Bariana, H., Jahier, J., Sivasithamparam, K., and Lagudah, E. S.
130–134. (2001). The introgressed segment carrying rust resistance genes Yr17, Lr37
Kruijer, W., Boer, M. P., Malosetti, M., Flood, P. J., Engel, B., Kooke, R., et al. (2015). and Sr38 in wheat can be assayed by a cloned disease resistance gene-
Marker-based estimation of heritability in immortal populations. Genetics 199, like sequence. Theor. Appl. Genet. 102, 600–605. doi: 10.1007/s00122005
379–398. doi: 10.1534/genetics.114.167916 1686
Malaker, P. K., Barma, N. C. D., Tiwari, T. P., Collis, W. J., Duveiller, E., Singh, P. K., Sehgal, D., Rosyara, U., Mondal, S., Singh, R., Poland, J., and Dreisigacker, S.
et al. (2016). First report of wheat blast caused by Magnaporthe oryzae pathotype (2020). Incorporating genome-wide association mapping results into genomic
Triticum in Bangladesh. Plant Dis. 100, 2330–2330. doi: 10.1094/pdis-05-16- prediction models for grain yield and yield stability in CIMMYT spring bread
0666-pdn wheat. Front. Plant Sci. 11:197. doi: 10.3389/fpls.2020.00197
Meuwissen, T. H. E., Hayes, B. J., and Goddard, M. E. (2001). Prediction of Singh, P. K., Gahtyari, N. C., Roy, C., Roy, K. K., He, X., Tembo, B., et al.
total genetic value using genome-wide dense marker maps. Genetics 157, (2021). Wheat Blast: A disease spreading by intercontinental jumps and its
1819–1829. management strategies. Front. Plant Sci. 12:710707. doi: 10.3389/fpls.2021.
Money, D., Gardner, K., Migicovsky, Z., Schwaninger, H., Zhong, G., and Myles, 710707
S. (2015). LinkImpute: fast and accurate genotype imputation for nonmodel Singh, R. P., Singh, P. K., Rutkoski, J., Hodson, D. P., He, X., Jørgensen, L. N.,
organisms. G3 Genes Genomes Genet. 5, 2383–2390. doi: 10.1534/g3.115.02 et al. (2016). Disease impact on wheat yield potential and prospects of genetic
1667 control. Annu. Rev. Phytopathol. 54, 303–322. doi: 10.1146/annurev-phyto-
Mottaleb, K. A., Singh, P. K., Sonder, K., Kruseman, G., and Erenstein, O. (2019b). 080615-095835
Averting wheat blast by implementing a ‘wheat holiday’: In search of alternative Spindel, J., Begum, H., Akdemir, D., Virk, P., Collard, B., Redona, E., et al. (2015).
crops in West Bengal, India. PLoS One 14:e0211410. doi: 10.1371/journal.pone. Genomic selection and association mapping in rice (Oryza sativa): Effect of
0211410 trait genetic architecture, training population composition, marker number and
Mottaleb, K. A., Singh, P. K., He, X., Hossain, A., Kruseman, G., and Erenstein, O. statistical model on accuracy of rice genomic selection in elite, tropical rice
(2019a). Alternative use of wheat land to implement a potential wheat holiday breeding line. PLoS Genet. 11:e1004982. doi: 10.1371/journal.pgen.1004982
as wheat blast control: In search of feasible crops in Bangladesh. Land Use Policy Tagle, A. G., Chuma, I., and Tosa, Y. (2015). Rmg7, a new gene for resistance
82, 1–12. doi: 10.1016/j.landusepol.2018.11.046 to Triticum isolates of Pyricularia oryzae identified in tetraploid wheat.
Mottaleb, K. A., Singh, P. K., Sonder, K., Kruseman, G., Tiwari, T. P., Barma, Phytopathology 105, 495–499. doi: 10.1094/PHYTO-06-14-0182-R
N. C. D., et al. (2018). Threat of wheat blast to South Asia’s food security: Takabayashi, N., Tosa, Y., Oh, H. S., and Mayama, S. (2002). A gene-for-
An ex-ante analysis. PLoS One 13:e0197555. doi: 10.1371/journal.pone.019 gene relationship underlying the species-specific parasitism of Avena/Triticum
7555 isolates of Magnaporthe grisea on wheat cultivars. Phytopathology 92, 1182–
Odilbekov, F., Armoniené, R., Koc, A., Svensson, J., and Chawade, A. (2019). 1188. doi: 10.1094/PHYTO.2002.92.11.1182
GWAS-assisted genomic prediction to predict resistance to Septoria tritici Tembo, B., Mulenga, R. M., Sichilima, S., M’siska, K. K., Mwale, M., Chikoti,
blotch in Nordic winter wheat at seedling stage. Front. Genet. 10:1224. doi: P. C., et al. (2020). Detection and characterization of fungus (Magnaporthe
10.3389/fgene.2019.01224 oryzae pathotype Triticum) causing wheat blast disease on rain-fed grown
Ornella, L., Sukhwinder-Singh, P., Perez, P., Burgueño, J., Singh, R., Tapia, wheat (Triticum aestivum L.) in Zambia. PLoS One 15:e0238724. doi: 10.1371/
E., et al. (2012). Genomic prediction of genetic values for resistance to journal.pone.0238724
wheat rusts. Plant Genome 5, 136–148. doi: 10.3835/plantgenome2012.07. Tosa, Y., and Chuma, I. (2014). Classification and parasitic specialization of blast
0017 fungi. J. Gen. Plant Pathol. 80, 202–209. doi: 10.1007/s10327-014-0513-7
Perelló, A., Martinez, I., and Molina, M. (2015). First report of virulence and effects Urashima, A. S., Grosso, C. R. F., Stabili, A., Freitas, E. G., Silva, C. P., Netto,
of Magnaporthe oryzae isolates causing wheat blast in Argentina. Plant Dis. D. C. S., et al. (2009). “Effect of Magnaporthe grisea on seed germination, yield
99:1177. doi: 10.1094/PDIS-11-14-1182-PDN and quality of wheat,” in Advances in Genetics, Genomics and Control of Rice
Pérez, P., and de los Campos, G. (2014). Genome-wide regression and prediction Blast Disease, eds G. L. Wang and B. Valent (Dordrecht: Springer).
with the BGLR statistical package. Genetics 198, 483–495. doi: 10.1534/genetics. Valent, B., Cruppe, G., Stack, J. P., Cruz, C. D., Farman, M. L., Paul, P. A., et al.
114.164442 (2021). Recovery plan for wheat blast caused by Magnaporthe oryzae pathotype
Picinini, E. C., and Fernandes, J. M. C. (1990). Occurrence of wheat blast Triticum. Plant Health Prog. 22, 182–212. doi: 10.1094/php-11-20-0101-rp
(Pyricularia oryzae) in commercial fields in the state of Rio Grande do Sul, van Ginkel, M., and Rajaram, S. (1993). “Breeding for durable resistance to diseases
Brazil. Fitopatol. Bras. 15, 83–84. in wheat an additional perspective,” in Durability of Disease Resistance. Current
Poland, J. A., Brown, P. J., Sorrells, M. E., and Jannink, J. L. (2012). Development Plant Science and Biotechnology in Agriculture, eds T. Jacobs and J. E. Parlevliet
of high-density genetic maps for barley and wheat using a novel two-enzyme (Dordrecht: Springer), 259–272.
genotyping-by-sequencing approach. PLoS One 7:e32253. doi: 10.1371/journal. VanRaden, P. M. (2008). Efficient methods to compute genomic predictions.
pone.0032253 J. Dairy Sci. 91, 4414–4423. doi: 10.3168/jds.2007-0980
Rajaram, S., Van Ginkel, M., and Fisher, R. A. (1993). “CIMMYT’s wheat breeding Viedma, L. Q., and Morel, W. (2002). Añublo o Piricularia del Trigo. Díptico.
mega-environments (ME),” in Proceedings of the 8th International Wheat Itapúa: MAG/DIA/CRIA.
Genetic Symposium, Tulln, 1101–1106. Voss-Fels, K. P., Cooper, M., and Hayes, B. J. (2019). Accelerating crop genetic
Roy, K. K., Anwar, M. B., Mustarin, K. E., Reza, M. M. A., Muzahid-E-Rahman, gains with genomic selection. Theor. Appl. Genet. 132, 669–686. doi: 10.1007/
M., Malaker, P. K., et al. (2021). Evaluation of different fungicides (chemical, s00122-018-3270-8
Frontiers in Plant Science | www.frontiersin.org 17 January 2022 | Volume 12 | Article 745379
fpls-12-745379 January 3, 2022 Time: 12:58 # 18
Juliana et al. Genomic Selection for Wheat Blast
Wang, S., Asuke, S., Vy, T. T. P., Inoue, Y., Chuma, I., Win, J., et al. Zhang, N., Luo, J., Rossman, A. Y., Aoki, T., Chuma, I., Crous, P. W., et al.
(2018). A new resistance gene in combination with Rmg8 confers strong (2016). Generic names in Magnaporthales. IMA Fungus 7, 155–159. doi: 10.
resistance against Triticum isolates of Pyricularia oryzae in a common 5598/imafungus.2016.07.01.09
wheat landrace. Phytopathology 108, 1299–1306. doi: 10.1094/PHYTO-12-17-
0400-R Conflict of Interest: The authors declare that the research was conducted in the
Wang, Y., Zhang, H., Xie, J., Guo, B., Chen, Y., Zhang, H., et al. (2018). Mapping absence of any commercial or financial relationships that could be construed as a
stripe rust resistance genes by BSR-Seq: YrMM58 and YrHY1 on chromosome potential conflict of interest.
2AS in Chinese wheat lines Mengmai 58 and Huaiyang 1 are Yr17. Crop J. 43,
323–331. doi: 10.1016/j.cj.2017.03.002 Publisher’s Note: All claims expressed in this article are solely those of the authors
Wickham, H. (2009). ggplot2: Elegant Graphics for Data Analysis. Berlin: Springer. and do not necessarily represent those of their affiliated organizations, or those of
Wu, L., He, X., Kabir, M. R., Roy, K. K., Anwar, M. B., Marza, F., et al. (2021). the publisher, the editors and the reviewers. Any product that may be evaluated in
Genetic sources and loci for wheat head blast resistance identified by genome- this article, or claim that may be made by its manufacturer, is not guaranteed or
wide association analysis. Crop J. doi: 10.1016/j.cj.2021.07.007 [Epub ahead of endorsed by the publisher.
print].
Xue, S., Kolmer, J. A., Wang, S., and Yan, L. (2018). Mapping of leaf rust resistance Copyright © 2022 Juliana, He, Marza, Islam, Anwar, Poland, Shrestha, Singh,
genes and molecular characterization of the 2NS/2AS translocation in the wheat Chawade, Joshi, Singh and Singh. This is an open-access article distributed under the
cultivar Jagger. G3 Genes Genomes Genet. 8, 2059–2065. doi: 10.1534/g3.118. terms of the Creative Commons Attribution License (CC BY). The use, distribution
200058 or reproduction in other forums is permitted, provided the original author(s) and
Zhan, S. W., Mayama, S., and Tosa, Y. (2008). Identification of two genes for the copyright owner(s) are credited and that the original publication in this journal
resistance to Triticum isolates of Magnaporthe oryzae in wheat. Genome 51, is cited, in accordance with accepted academic practice. No use, distribution or
216–221. doi: 10.1139/G07-094 reproduction is permitted which does not comply with these terms.
Frontiers in Plant Science | www.frontiersin.org 18 January 2022 | Volume 12 | Article 745379