Edited by: Jacqueline Batley, University of Western Australia, Australia
Reviewed by: Wenxin Liu, China Agricultural University (CAU), China; Reif Jochen, Leibniz Institute of Plant Genetics and Crop Plant Research (IPK), Germany
*Correspondence: Yoseph Beyene,
This article was submitted to Plant Breeding, a section of the journal Frontiers in Plant Science
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
Genomic selection predicts the genomic estimated breeding values (GEBVs) of individuals not previously phenotyped. Several studies have investigated the accuracy of genomic predictions in maize but there is little empirical evidence on the practical performance of lines selected based on phenotype in comparison with those selected solely on GEBVs in advanced testcross yield trials. The main objectives of this study were to (1) empirically compare the performance of tropical maize hybrids selected through phenotypic selection (PS) and genomic selection (GS) under well-watered (WW) and managed drought stress (WS) conditions in Kenya, and (2) compare the cost–benefit analysis of GS and PS. For this study, we used two experimental maize data sets (stage I and stage II yield trials). The stage I data set consisted of 1492 doubled haploid (DH) lines genotyped with rAmpSeq SNPs. A subset of these lines (855) representing various DH populations within the stage I cohort was crossed with an individual single-cross tester chosen to complement each population. These testcross hybrids were evaluated in replicated trials under WW and WS conditions for grain yield and other agronomic traits, while the remaining 637 DH lines were predicted using the 855 lines as a training set. The second data set (stage II) consists of 348 DH lines from the first data set. Among these 348 best DH lines, 172 lines selected were solely based on GEBVs, and 176 lines were selected based on phenotypic performance. Each of the 348 DH lines were crossed with three common testers from complementary heterotic groups, and the resulting 1042 testcross hybrids and six commercial checks were evaluated in four to five WW locations and one WS condition in Kenya. For stage I trials, the cross-validated prediction accuracy for grain yield was 0.67 and 0.65 under WW and WS conditions, respectively. We found similar responses to selection using PS and GS for grain yield other agronomic traits under WW and WS conditions. The top 15% of hybrids advanced through GS and PS gave 21%–23% higher grain yield under WW and 51%–52% more grain yield under WS than the mean of the checks. The GS reduced the cost by 32% over the PS with similar selection gains. We concluded that the use of GS for yield under WW and WS conditions in maize can produce selection candidates with similar performance as those generated from conventional PS, but at a lower cost, and therefore, should be incorporated into maize breeding pipelines to increase breeding program efficiency.
With more than 35 million ha harvested each year, maize is the most important staple food crop in sub-Saharan Africa (SSA). In SSA countries, maize is commonly grown by resource-poor farmers and covers large areas with very low average grain yield (1.4 ton/ha) (
Genomic prediction is an approach that uses molecular marker data to predict the genetic value of complex traits in progeny for selection and breeding (
As pointed out by
Although testing predictive ability is critical for gathering information for GS, there is a large gap between the findings of these studies and their application in breeding programs (
The current study compares the performance of maize DH line testcrosses selected based on GS versus PS in second stage multi-location yield trials of the CIMMYT maize breeding program in SSA. For this study, we used two experimental maize data sets: first-stage multi-location yield trials (hereafter referred to as stage I) and second-stage multi-location yield trials (hereafter referred to as stage II). The stage I data set consisted of 1492 DH lines genotyped with rAmpSeq (epeat lification uencing) dominant sequence tag markers (
The first data set (stage I) comprised a total of 1492 DH lines derived from 12 bi-parental DH populations developed at CIMMYT’s Maize DH facility in Kiboko, Kenya. The 12 source populations were obtained by crossing elite CIMMYT maize lines (CMLs) with La Posta Seq C7, a drought tolerant population developed at CIMMYT, Mexico, through recurrent selection among full sib/S1 families (
List of 12 bi-parental maize populations used in this study.
No. | Population name | # Doubled haploid (DH) lines genotyped | # DH lines phenotyped |
---|---|---|---|
1 | CML440/LPS-F64 | 34 | 34 |
2 | CML445/LPS-F64 | 181 | 91 |
3 | CML312/LPS-F64 | 185 | 93 |
4 | CML442/LPS-F64 | 240 | 126 |
5 | CML505/LPS-F64 | 162 | 81 |
6 | CZL04003/LPS-F64 | 134 | 67 |
7 | CML536/LPS-F64 | 180 | 86 |
8 | CML537/LPS-F64 | 110 | 55 |
9 | CML538/LPS-F64 | 40 | 40 |
10 | CML540/LPS-F64 | 51 | 51 |
11 | ZEWAc1F2-134-4-1-B-1-B*4-1-2-B-B/LPS-F64 | 75 | 75 |
12 | CML312/CML540 | 100 | 52 |
|
|
|
To implement GS in CIMMYT’s maize breeding program, nearly half (855) of 1492 selected DH lines were crossed with a single-cross tester from complementary heterotic group and phenotyped across locations. The 855 hybrids were divided into 14 trials connected by common checks. In each trial, three to six commercial checks were included and planted in an alpha-lattice design with two replications and phenotyped in three well-watered (WW) environments and one managed drought stress (WS) environment in Kenya during the 2017 growing season. The WS experiment was conducted during the dry (rain-free) season by suspending irrigation starting 2 weeks before flowering until harvest, whereas the WW experiments were conducted during the rainy season, applying supplemental irrigation as needed. Entries were planted in two-row plots, 5 m long, with 0.75 m spacing between rows and 0.25 m between hills. Two seeds per hill were initially planted and then thinned down to one plant per hill three weeks after emergence to obtain a final plant population density of 53,333 plants per hectare. Fertilizers were applied at the rate of 60 kg N and 60 kg P2O5 per ha, as recommended for the area. Nitrogen was applied twice: at planting and 6 weeks after emergence. Fields were kept free of weeds by hand weeding. The following traits were measured: grain yield (GY, tons ha− 1), anthesis date (AD, days), plant height (PH, cm), grain moisture (MOI, %), gray leaf spot (GLS, 1–5 rating score), and turcicum leaf blight (TLB, 1–5 rating score). Plots were manually harvested and GY was corrected to 12.5% moisture. AD was measured from planting to when 50% of the plants shed pollen, and PH was measured from the soil surface to the flag leaf collar on five representative plants within each plot.
Leaf samples were taken from each of the 1492 DH lines and sent to Intertek, Sweden, for DNA extraction. The DNA sample plates were forwarded to the Institute for Genomic Diversity, Cornell University, Ithaca, NY, USA, for genotyping with repetitive sequences (rAmpSeq markers) as per the procedure described by
From stage I analyses, the top performing 348 (23%) DH lines were chosen for stage II evaluation. Among these 348 DH lines, 172 lines represented selection from the 637 genomic predicted lines that had above average GEBVs and 176 lines were selected from the 855 phenotyped lines that had above average Best Linear Unbiased Estimates (BLUE). Each of these DH lines were crossed with three common testers from complementary heterotic groups. The resultant 1042 testcross hybrids were evaluated in eight connected trials. Six commercial checks were included in each trial and planted in an alpha-lattice design with two replications and phenotyped in 4-5 WW environments and one WS environment in Kenya in 2018. The WS experiment was conducted during the dry (rain-free) season by suspending irrigation starting 2 weeks before flowering until harvest, whereas the WW experiments were conducted during the rainy season, applying supplemental irrigation as needed. Planting and agronomic managements were similar as explained for stage I trials. The following traits were measured: grain yield (GY, tons ha−1), anthesis date (AD, days), plant height (PH, cm), grain moisture (MOI, %), gray leaf spot (GLS, 1–5 rating score), and turcicum leaf blight (TLB, 1–5 rating score). Plots were manually harvested and GY was corrected to 12.5% moisture.
There were two sets of phenotypic field trials; the first set included 855 hybrids used to predict the performance of unobserved 637 lines (stage I), and the second set (stage II) was made up of 1042 hybrids from 348 DH lines (172 lines selected from GEBV alone and 176 lines selected based on phenotypic data) crossed with three testers. Note that the second set of field trials was used to compare the performance of the GS vs PS of hybrids. All the phenotypic analyses were done to obtain the variance components and BLUEs for the lines under WW and WS. All testcrosses were evaluated in different trials but adjacent to each other and connected by common checks in the same field. Phenotypic data was analyzed first within trials and then across trials.
The BLUEs across WW and WS locations for each trial and each trait were generated using the following linear mixed model carried out using the META-R software (
where
The analysis across trials was also performed using similar model as those shown above but including the trial as fixed effect.
The BLUE of the entries within and across testers were used for genome-based predictions. GEBVs were calculated for GY, AD, MOI and PH using the BGLR statistical R-package (
The models described below were used with two purposes: one was to use the 855 lines as a training set to predict the GEBV of 637 lines (testing set) and use the observed and predicted values to select top performing lines. The other objective of the models described below was to study the genome-based prediction accuracy of the 855 lines with phenotypic and genotypic data and determine the prediction accuracy using main effects and main effects plus interaction models for each tester and across testers.
This model can be expressed as
Where
Note that this model was used for the genomic prediction computed for the WW sites. The predictions for the unique managed sites had only the G + e component because these trials were established in only one managed drought site.
This is the same as the previous model but includes the interaction term based on marker and environment interaction data. The model (
Where
The performance of the models when predicting the five traits was evaluated using the average Pearson’s correlation coefficient between observed and predicted values. The random cross-validation scheme mimics real plant breeding situations and is a scheme where the performance of 20% of the maize testcrosses was not observed in any of the environments and the rest of the lines (80%) were already observed in the same target environments. For this scheme, a five-fold random partitioning (80% of the data used as the training set, and the remaining 20% as the testing set) was employed. Four folds were used for training the models and for predicting the remaining fold. This procedure was repeated over the five folds and the predictions from the testing fold were joined in a single vector. Then, Pearson’s correlations between predicted and observed values within the same environment were computed. The partitioning was repeated 100 times. The cost benefits of PS vs GS were analyzed using spreadsheet-based budgeting tools.
For stage I, mean GY averaged across WW locations ranged from 3.49 to 9.14 t/ha with an overall mean of 6.03 t/ha, whereas at stage II it improved further, ranging from 5.1 to 11.6 t/ha with an average of 7.59 t/ha (
Mean, range, genetic variance, and broad-sense heritability estimates for grain yield (GY, t/ha) anthesis date (AD, days), plant height (PH, cm), moisture (MOI, %), gray leaf spot (GLS, 1–5 rating score), and turcicum leaf blight (TLB, 1–5 rating score) for stage I and stage II testcrosses evaluated under optimum and managed drought stress conditions in Kenya.
Optimum (Stage I) | Managed drought (Stage I) | |||||||||
---|---|---|---|---|---|---|---|---|---|---|
GY | AD | PH | MOI | GLS | TLB | GY | AD | PH | MOI | |
Mean | 6.03 | 64.31 | 235.56 | 16.81 | 2.31 | 3.43 | 3.25 | 63.3 | 207.7 | 16.2 |
Min | 3.49 | 53.71 | 194.33 | 14.13 | 0.81 | 1.97 | 1.08 | 57.98 | 163.53 | 8.05 |
Max | 9.14 | 73.90 | 270.04 | 21.60 | 4.18 | 5.04 | 5.76 | 69.88 | 244.65 | 23.20 |
Checks Mean | 6.14 | 64.89 | 246.25 | 16.08 | 2.40 | 2.90 | 2.99 | 63.90 | 222.69 | 15.96 |
σ2 G | 0.19** | 1.29** | 41.69** | 0.18** | 0.01* | 0.10** | 0.17** | 1.96** | 49.82** | 0.91** |
σ2 T | 0.00 | 0.90 | 11.59 | 0.36 | 0.02 | 0.07 | 0.52 | 0.74 | 0.63 | 0.43 |
σ2 E | 0.42 | 145.11 | 1200.50 | 14.70 | 0.00 | 0.00 | – | – | – | – |
σ2 GxE | 0.18** | 0.14** | 0.06* | 0.00 | 0.01* | 0.10** | – | – | – | – |
σ2 GxT | 0.74** | 3.00** | 81.18** | 0.81** | 0.02** | 0.07** | – | – | – | – |
σ2 e | 1.41 | 2.82 | 160.47 | 3.19 | 0.16 | 0.24 | 0.33 | 1.36 | 59.08 | 2.43 |
|
0.46 | 0.77 | 0.67 | 0.31 | 0.37 | 0.65 | 0.88 | 0.95 | 0.92 | 0.84 |
LSD | 2.09 | 4.11 | 21.15 | 2.52 | 0.89 | 1.00 | 1.66 | 2.50 | 24.30 | 3.40 |
CV | 19.66 | 2.61 | 5.38 | 10.62 | 17.31 | 14.12 | 17.56 | 1.80 | 3.70 | 9.70 |
Optimum (Stage II) | Managed drought (Stage II) | |||||||||
Mean | 7.59 | 71.9 | 247.6 | 19.64 | 1.98 | 2.58 | 3.23 | 72.40 | 206.40 | 14.60 |
Min | 5.10 | 62.0 | 196.7 | 16.54 | 1.43 | 1.50 | 0.77 | 63.50 | 159.60 | 9.70 |
Max | 11.67 | 77.5 | 291.2 | 22.22 | 4.45 | 4.08 | 6.33 | 81.00 | 247.10 | 22.90 |
Checks Mean | 6.90 | 69.43 | 262.94 | 18.83 | 2.03 | 1.76 | 2.31 | 71.80 | 217.50 | 14.40 |
σ2 G | 0.33** | 2.6** | 101.0** | 0.24** | 0.00 | 0.09** | 0.20** | 2.30** | 73.50** | 1.00** |
σ2 T | 0.37 | 1.9 | 97.1 | 0.00 | 0.00 | 0.08 | 0.55 | 7.1 | 44.10 | 4.60 |
σ2 E | 2.06 | 75.8 | 389.5 | 10.03 | 0.00 | 0.01 | – | – | – | – |
σ2 GxE | 0.25** | 1.10** | 189.3** | 2.31** | 0.04 | 0.00 | – | – | – | – |
σ2 GxT | 0.23** | 0.20* | 10.9** | 0.21** | 0.01 | 0.00 | – | – | – | – |
σ2 e | 1.27 | 2.10 | 102.3 | 4.38 | 0.10 | 0.19 | 0.40 | 1.40 | 68.90 | 2.10 |
|
0.38 | 0.14 | 0.53 | 0.09 | 0.82 | 0.18 | 0.34 | 0.32 | 0.61 | 0.52 |
LSD | 2.11 | 2.00 | 29.30 | 4.48 | 0.63 | 0.66 | 1.42 | 2.70 | 21.10 | 3.00 |
CV | 14.82 | 2.00 | 4.10 | 10.66 | 16.20 | 16.83 | 19.93 | 1.70 | 4.00 | 9.9 |
*,** Significance at P < 0.05 and 0.01 level, respectively.
The cross-validation analyses yielded moderately high prediction correlations among optimum and drought conditions for GY and other agronomic traits. The prediction correlations ranged from 0.65–0.67 for GY, 0.57–0.65 for MOI, 0.67–0.75 for AD, and 0.70–0.72 for PH (
Prediction accuracy for each tester and across testers under cross-validation scenarios for grain yield (GY), anthesis date (AD), plant height (PH), and moisture content (MOI) evaluated under well-watered (WW) and water stress (WS) conditions in Kenya.
Trait | Model\Tester | Within tester | Across testers | ||
---|---|---|---|---|---|
CML312 × CML395 | CML312 × CML442 | CML395 × CML444 | |||
Total Hybrids | 111 | 742 | 979 | ||
GY-WW | G | 0.41 ± 0.09 | 0.16 ± 0.12 | 0.60 ± 0.03 | 0.67 ± 0.05 |
G + GE | 0.42 ± 0.07 | 0.19 ± 0.07 | 0.59 ± 0.04 | – | |
GY-WS | G | 0.75 ± 0.04 | 0.22 ± 0.18 | 0.64 ± 0.07 | 0.65 ± 0.05 |
MOI-WW | G | 0.58 ± 0.05 | 0.16 ± 0.14 | 0.58 ± 0.01 | 0.65 ± 0.04 |
G + GE | 0.61 ± 0.07 | 0.09 ± 0.07 | 0.59 ± 0.04 | – | |
MOI- WS | G | 0.09 ± 0.04 | 0.44 ± 0.16 | 0.61 ± 0.06 | 0.57 ± 0.05 |
AD-WW | G | 0.58 ± 0.07 | 0.41 ± 0.13 | 0.70 ± 0.07 | 0.75 ± 0.04 |
G + GE | 0.53 ± 0.19 | 0.40 ± 0.14 | 0.74 ± 0.04 | – | |
AD- WS | G | 0.51 ± 0.06 | 0.49 ± 0.20 | 0.63 ± 0.04 | 0.67 ± 0.05 |
PH-WW | G | 0.28 ± 0.10 | 0.14 ± 0.12 | 0.65 ± 0.03 | 0.70 ± 0.03 |
G + GE | 0.40 ± 0.06 | 0.16 ± 0.09 | 0.67 ± 0.03 | – | |
PH-WS | G | 0.52 ± 0.03 | 0.17 ± 0.12 | 0.72 ± 0.04 | 0.72 ± 0.04 |
A total 172 lines that had above average GEBVs were advanced to stage II evaluations (
At stage II, 1042 hybrids were evaluated among them 526 were developed from lines selected based on PS and the remaining were derived from lines selected based on GEBVs. The GY of 526 testcross hybrids advanced through PS evaluated across five WW locations (hereafter referred to as PS-WW) ranged from 5.54 to 11.67 t /ha (
Comparison of hybrids advanced through genomic and phenotypic selections and commercial checks evaluated at stage II or advanced yield trials under optimum and managed drought stress across Kenya in 2018.
Phenotypic selection (PS) | Genomic selection (GS) | |||
---|---|---|---|---|
Well-watered (GY, t/ha) | Water stress (GY, t/ha) | Well-watered (GY, t/ha) | Water stress (GY, t/ha) | |
All hybrids | 7.7 | 3.2 | 7.5 | 3.2 |
Top 15% of hybrids | 9.4 | 4.7 | 9.1 | 4.8 |
Best hybrid | 11.7 | 6.2 | 10.4 | 6.3 |
Mean of commercial checks | 7.2 | 2.3 | 7.2 | 2.3 |
Best check | 8.4 | 3.3 | 8.4 | 3.3 |
|
||||
Top 15% of hybrids over commercial checks | 23% | 51% | 21% | 52% |
Top 15% of hybrids over the best commercial check | 12% | 42% | 8% | 47% |
The best hybrid over commercial checks | 63% | 169% | 46% | 175% |
The best hybrid over the best commercial check | 39% | 88% | 24% | 92% |
A total of 516 hybrids advanced through GS and evaluated at the same five WW locations (hereafter referred to as GS-WW) produced GY ranging from 5.1 to 10.44 t/ha (
The top 15% of hybrids advanced through PS on average had an increase of 8.5 cm in PH and 4.7 days in AD and a 1% increase in grain moisture content compared to the mean of the commercial checks (
A total of 526 hybrids advanced through PS were also evaluated under managed drought stress (hereafter referred to as PS-WS); their mean GY ranged from 0.99 to 6.19 t ha−1 (
The best hybrid advanced through GS produced 92% and 175% higher GY than the best check and mean of commercial checks, respectively (
An additional metric of interest when considering the overall efficacy of GS as a substitute for conventional PS only schemes is the advancement rate of the GS stage II cohort compared with the advancement rate of the PS stage II cohort. The actual advancement rate of the two methods is a useful means of comparing the overall value of the two groups of advanced lines since it captures all information that the breeder uses to make the decision whether or not to move a given DH line/s into advanced testing (
Performance of hybrids advanced through genomic selection (GS), phenotypic selection (PS), and commercial checks evaluated in stage II trials under optimum conditions for grain yield (GY, t/ha), anthesis date (AD, days), plant height (PH, cm), and turcicum leaf blight (TLB, 1–5 rating score). The numbers in the bracket indicated the total number of hybrids.
Performance of hybrids advanced through GS and PS and commercial checks evaluated in stage II trials under managed drought for GY (t/ha), AD (days), and PH (cm). The numbers in the bracket indicated the total number of hybrids.
While under WS condition the top 15% (157 of 1042 hybrids) advanced to stage III trials, 91 hybrids were developed through GS, and 66 hybrids were advanced through PS. There was no significant difference among the top 15% of hybrids advanced through PS and GS for grain yield, AD, and PH (
We compared the costs involved in PS and GS using spreadsheet-based budgeting tools (
Cost–benefit analysis of phenotypic selection and genomic selection in International Maize and Wheat Improvement Center’s (CIMMYT’s) maize breeding program in Kenya.
Methods | Cost/entry (US$) | No. of entries | No. of reps/sites | No. of rows/sites | No. of sites | Total cost (US$) |
---|---|---|---|---|---|---|
PS (making testcrosses) | 10 | 1492 | 1 | 1 | 1 | 14,920 |
PS (stage I multi-location yield trials) | 5 | 1492 | 2 | 2 | 4 | 119,360 |
GS (making testcrosses) | 10 | 855 | 1 | 1 | 1 | 8,550 |
GS (phenotyping training set in stage I multi-location yield trials) | 5 | 855 | 2 | 2 | 4 | 68,400 |
GS (genotyping all lines) | 10 | 1492 | 1 | 1 | 1 | 14,920 |
Total cost of GS | 91,870 | |||||
Total cost of PS | 134,280 | |||||
GS:PS cost ratio | 0.68 |
With the advent of DH technology, thousands of fixed lines are generated each year in maize. However, identifying the best genotypes requires extensive field evaluations with several hybrid combinations, and all DH lines cannot be evaluated because of limited space and resources. One method for reducing the number of hybrids for field evaluation is crossing all DH lines with a common tester in the early stages of a breeding cycle. Another method is to use a genetic similarity matrix derived from pedigree or molecular markers for predicting performance of untested crosses (
In this study, we compared the performance of maize DH lines selected from stage I multi-location yield trials based on BLUEs and GEBVs by evaluating the hybrids in common stage II multi-location yield trials of the CIMMYT maize breeding program. We evaluated a total of 855 hybrids under optimum and drought conditions and used BLUEs data to predict the remaining 637 hybrids which were genotyped but have never been phenotyped. In our study, the prediction accuracy for GY under WW conditions was 0.67, and under WS, it was 0.65 (
Identification of optimum size as training and prediction set is crucial for implementing GS in maize breeding program.
The mean GY of hybrids advanced through GS and PS methods was significantly higher than the mean of the commercial checks (
Comparison of hybrids advanced through PS and GS under drought stress conditions revealed that GS did slightly better (4.68 t/ha was the mean of the top 15% of hybrids) than PS (4.48 t/ha, mean of the top 15% of hybrids). There was no significant difference among the top 15% of hybrids advanced through PS and GS for other traits.
GS was found to outperform MAS using the same financial investment, even at low prediction accuracies (
The largest potential advantage of GS is predicting the breeding value of genotyped parents that were never phenotyped. We found similar responses to selection using PS and GS for grain yield under WW and WS conditions. The top 15% of hybrids advanced through GS and PS produced 21% to 23% higher GY under WW and 52% to 51% under WS than the mean of the commercial checks. The GS reduced the cost by 32% over the PS with similar selection gains. We conclude that the use of GS for yield under optimum and drought conditions in tropical maize can produce selection candidates with similar performance as those generated from conventional PS, but at a lower cost; therefore, this strategy should be effectively incorporated into maize breeding pipelines to enhance breeding program efficiency.
All datasets generated for this study are included in the article/
YB, MG, MO, BP, JC, KR, and SM contributed in the project planning and overall coordination. YB performed and coordinated the field experiments. MG, SG, and KD were responsible for coordinating sample management and genotyping. JC, PP, KR, GA, MG, and YB carried out the analysis. YB, JC, and MG wrote the manuscript. All authors have made their contribution in editing the manuscript and approved the final version.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
This research was supported by the Bill and Melinda Gates Foundation, and the United States Agency for International Development (USAID) through Stress Tolerant Maize for Africa (STMA, Grant # OPP1134248), and the CGIAR Research Program MAIZE. The CGIAR Research Program MAIZE receives W1&W2 support from the Governments of Australia, Belgium, Canada, China, France, India, Japan, Korea, Mexico, Netherlands, New Zealand, Norway, Sweden, Switzerland, United Kingdom, United States, and the World Bank. We are grateful to CIMMYT Field Technicians at different stations in Kenya for data collection; CIMMYT Laboratory Technicians in Kenya for sample preparation for genotyping; and Cornell Genotyping facility and Buckler’s group for genotyping services and data turn-around.
The Supplementary Material for this article can be found online at: