Association Analysis and Genome-wide Selection for Early Maturity in Wheat. Thesis. Science in the Graduate School of The Ohio State University

Transcription

1 Association Analysis and Genome-wide Selection for Early Maturity in Wheat Thesis Presented in Partial Fulfillment of the Requirements for the Degree of Master of Science in the Graduate School of The Ohio State University By Nafeti Titus Mheni, B.S. Graduate Program in Horticulture and Crop Science The Ohio State University 2014 Thesis Committee, Clay H. Sneller, Advisor David Francis Leah McHale i

2 Copyrighted by Nafeti Titus Mheni 2014 i

3 Abstract Crop phenology is an important component of wheat adaptation to climate change. A few major genes and QTLs, along with minor genes have been reported to control variation for flowering and maturity in wheat. This study aimed to 1) identify QTLs for heading date (HD) 2) evaluate the accuracy and relative efficiency of genomic selection (GS) versus phenotypic selection for HD 3) assess the stability of alleles and genomic selection models for HD in spring and winter wheat. We used a soft winter wheat (SWW) panel phenotyped for HD in North America, and a hard spring wheat (HSW) panel phenotyped in the United States and Arusha Tanzania. The panels were genotyped with SNP markers. The analysis of genotype by environmental interaction produced two clusters of environments for each population with one cluster consisting of environments that produced a large range of HD and the other a narrow range. In both winter and spring wheat we detected seven very significant (p<0.0005) QTLs associated with HD. Within each population the QTL effects were consistent between clusters of environments. Very few QTLs were found to be common between the two populations, and the few that were in common had minor effects in either population. The accuracy and relative efficiency of GS was higher in the HSW population than the SWW population. We have identified nine genotypes that flower earlier than the existing commercial Tanzanian wheat varieties. The selected genotypes will be used as a resource in our breeding program for creating early flowering wheat varieties adapted to the Tanzanian environment. ii

4 Dedication I dedicate this work to my family, for their support for the past two years, without them this work might not have been written and to them I m greatly indebted. Furthermore I would like to thank my parents, my mother and father in law for their support to my family especially when I was away from home for studies. I would like to thank Dr. Clay Sneller and Prof. Susan Nchimbi-Msolla for their assistance and guidance for this work. iii

5 Acknowledgement I would like to express my special appreciation to my advisor. Dr. Clay Sneller, who have been wonderful advisor for me. I would like to thank you for encouragement and guidance that helped me to grow as a scientist. I take this opportunity to thank Prof. Susan Nchimbi-Msolla for her guidance, comments, suggestions and help. I would like to express my special thanks to the Innovative Agricultural Research Initiate (iagri) for their scholarship support and Norman Borlaug Leadership Enhancement in Agriculture program (LEAP) for funding part of my research work. I would like to thank my fellow graduate student Mao Huang for her assistance with data collection and support with analysis. I also want to thank the (TCAP) project for allowing me to use their wheat populations for my research. I would like to extend my thanks to Selian Agricultural Research Institute for their support when at the time when I was doing my research works while in Tanzania. I would like to thank Dr. David Francis and Dr. Leah McHale for serving as my committee members for their comments and suggestions, thank you all. Lastly, I want to thank all who in one way or another have helped me to accomplish this work. iv

6 Vita May Songea Boys Secondary School B. S. Agronomy, Sokoine University of Agriculture 2012 to Present Graduate Student, Department of Horticulture and Crop, The Ohio State University Horticulture and Crop Science Field of Study v

7 Table of Content Abstract i Dedication... iii Acknowledgement... iiv Vita... v List of Tables... viii List of Figures... iix 1. Introduction Literature Review Association Analysis and Genomic Selection Association analysis (AA) Genomic selection (GS) Materials and Methods Soft Red Winter Wheat Hard Spring Wheat Population Heritability and GEI analysis Genotyping Association Analyses and Genomic Selection Results SWW phenotypes and GEI HSW phenotypes and GEI Association analyses Genomic selection Discussion References Appendix A: Supplementary Table 1. Entries in the Soft Winter Wheat population and their BLUPs for heading date (days from planting) over all environments, environments in the North cluster, and environments in the South cluster vi

8 Appendix B: Supplementary Table 2. Entries in the Hard Spring Wheat population and their BLUPs for heading date (days from planting) over all environments, environments in the CATA cluster, and environments in the NA cluster vii

9 List of Tables Table 1. Summary of the Soft Winter Wheat growing environments, their main effect on heading date, and their cluster assignment Table 2. Summary of the Hard Spring Wheat growing environments, their main effect on heading date, and their cluster assignment Table 3. Mean squares, F-tests, variance components, and heritability of heading date for Soft winter wheat Table 4. Mean squares, F-tests, variance components, and heritability of heading date for Hard spring wheat Table 5. Total number of markers and significant markers for heading data Table 6. Summary of marker information for the Soft Winter Wheat (SWW) and Hard Spring Wheat populations Table 7. Summary of highly significant markers for heading date in the Soft Winter Wheat population 33 Table 8. Summary of highly significant markers for heading date in the Hard Spring Wheat population Table 9. Correlation of genomic estimated breeding values from 10 fold cross validation and relative efficiency of genomic selection viii

10 List of Figures Figure 1. Clustering of the Soft Winter Wheat environments using the matrix of genotype x environment interaction values and Ward s minimum variance criteria Figure 2. Bi-plot the main effect of Soft Winter Wheat environments and genotypes versus their scores for the first principal component from the AMMI analysis Figure 3. Main effect of Soft Winter Wheat genotypes versus average GEI of genotype with South environments Figure 4. Main effects of soft winter wheat genotypes versus average GEI of genotype with North Environment Figure 5. Clustering of the Hard Spring Wheat environments using the matrix of genotype x environment interaction values and Ward s minimum variance criteria Figure 6. Bi-plot the main effect of Hard Spring Wheat environments and genotypes versus their scores for the first principal component from the AMMI analysis Figure 7. Main effect of Hard Spring Wheat genotypes versus average GEI of genotype with California + Tanzanian environments Figure 8. Main effect of Hard Spring Wheat genotypes versus average GEI of genotype with North American environments ix

11 1. Introduction Wheat (Triticum aestivum, L.) is a staple grain worldwide and has contributed immensely to changing human civilization. The area under wheat production is larger than the areas used for production of any other commercial grain. Wheat is widely grown in the world and about 225 million hectares of farmland is used for wheat production (FAO STAT 2009). Wheat is adapted to diverse climatic conditions including environments with limited moisture.. Global wheat consumption is increasing and in sub-saharan Africa with consumption is increasing faster than any other cereal grain. There are a number of factors behind this increased wheat consumption in Africa including a growing population, increase in income, increased urbanization, wheat food aid, as well as increasing participation of women in the labor force. There has been a growing gap between domestic wheat production and consumption in Africa over the last several decades. In the 1990 s annual wheat imports averaged 18 million tons and imports rose to 27 million tons in the 2000 s (Shiferaw et al., 2011). About 81% of wheat consumed in developing countries is produced and consumed in the same areas (CIMMYT, 2005). Wheat demand in developing countries is expected to increase at a rate of 1.6 % per year by 2020 (Ortiz et al., 2008) African countries must import wheat from abroad to bridge the production gap. At the same time, African countries have large fertile land areas that can be used for wheat production, though most of that area is not currently used for wheat. There are a number of limiting factors for wheat production such as dependence on rain fed agriculture. In many parts of sub-saharan Africa small scale farmers are poorly equipped for wheat production yet are producing wheat, leading to environmental related production constraints Africa is the region of the world that may be most susceptible to the impacts of climate unpredictability and variation (Challinor et al., 2007). Climate change is increasing the frequency, duration, 1

12 and timing of drought and heat stress in Africa. Drought limits the growth of plants by inhibiting photosynthesis and reducing carbohydrate synthesis. Drought can cause wheat yield reduction through ovule abortion, pollen sterility, shrived seeds and kernels abortion. Global food shortage is in part a consequence of drought and heat stress, which results in increased global food prices. Floukes et al. (2007) estimated the average yield of wheat under normal conditions of water supply to be 8t/ha, but in the case of drought, especially after crop anthesis, severe yield reductions can occur. Climate change could cause more wheat yield reduction and may impact crop production and quality. In developing countries a greater effect is anticipated than in developed countries (Ceccarelli et al., 2010). Development of new wheat cultivars should take into account the potential effects of drought and heat stress. Given that plant breeding has a record of success in developing high yielding cultivars, then it should be possible to develop new cultivars with traits facilitating adaptation to climate change (Ceccarelli et al., 2010). Marker assisted selection (MAS) or breeding refers to the use of genetic fingerprints to identify superior plants. Marker assisted selection uses molecular markers in linkage disequilibrium with genes controlling a trait including quantitative trait loci (QTL). Marker assisted selection predicts the value of individual genes and the value of the individuals with those genes. Marker assisted selection is mostly useful for genes with a large effect on a trait. Genomic selection (GS) is similar to MAS but uses 1,000s markers scattered about the genome to predict breeding values across the entire genome. Genomic selection predicts the value of the entire genome of an individual making selection possible without phenotyping and is useful for traits controlled by many genes with small effects. In addition, the use of molecular markers provides a means for evaluating levels of genetic diversity for a crop. Understanding crop genetic diversity is an important step for its improvement, and provides the foundation for selection of superior parental combinations (Landjeva et al., 2007). The timing of flowering, grain filling, and maturity are important phenology stages for cereal crops. Grain yield can be severely reduced if these development stages occur when there is drought or heat stress. Wheat plants with a shorter growing period or season may have the ability to escape the effect of drought that occur toward the end of the growing season as experienced in the wheat growing areas of Africa (Araus et al., 2008). 2

13 The overall objective of this study is to improve wheat productivity through selection of cultivars with phenology adapted to climate change. The specific objectives of the study were: 1. To identify QTLs for heading date in winter and spring wheat. 2. To evaluate the relative efficiency of genomic selection versus phenotypic selection for heading date. 3. To assess the stability of QTL allele effects and genomic selection models for heading date over environments and types of wheat. 3

14 2. Literature Review Abiotic and biotic stresses limit crop production and productivity. Drought and high temperatures are the two major limiting factors for crop production. The process of grain filling in wheat is very sensitive to environmental conditions, and successful flowering in most plants produces high yield (Barnabás, et al., 2008). The optimum air temperature range that is ideal for spring wheat growth and development is about C (Hossain and Da Silva, 2012). Drought and high temperatures occurring at the flowering stage have been found to cause flower abortion and kernel shrinkage which can lower yield (Shah and Paulsen, 2003). Climate change represents an important risk to African agricultural production systems and farmer s income. Certainly agriculture will have to adjust dramatically to accommodate the anticipated changes in order to meet future food demands for the rapidly growing population (Müller, 2011). The majority of the world s poor people live in areas that are resource poor and risk prone. In many countries and especially in sub-saharan Africa, low-income live in marginal areas and agriculture is their main way of living. Having so many poor people living in marginal areas leaves the population vulnerable to risks associated with climate change. In marginal areas there is a large number of undernourished people as a consequence of low crop yields due to variable weather and many other factors. The number of undernourished people is predicted to increase in sub-saharan Africa as a consequence of climate change (Müller, 2011). Thus, there is a need for agriculture in this region to develop cropping systems that are adapted to new conditions resulting from climate change. Wheat (Triticum aestivum, L.) is considered to have high adaptation potential to diverse environmental situations (Sourdille et al., 2000). The wide adaptability of wheat is partly due to genetic diversity in flowering time and maturity (Lewis et al., 2008). There are two major categories of wheat classified according to growing temperatures, winter wheat and spring wheat. The difference between the two major categories are based on their vernalization requirement. Winter wheat requires exposure to a prolonged period of cold temperature (4-8 weeks of cold treatment) for floral primordial development, 4

15 while spring wheat does not need a vernalization period in order to complete its life cycle. Within both winter and spring wheat types, photoperiod and temperature are the major factors affecting plant growth rate and time to flowering (Lewis et al., 2008). In other cereal grains like rice, heading date (HD) is also one of the critical factors for adaptation to growing conditions. As a result, heading or flowering time in rice has been an important trait (Cao et al., 2010). Heading date in rice is mainly determined by photoperiod sensitivity and several genes are believed to be involved in controlling photoperiod sensitivity. In the highest latitude areas (approximately 45 N) rice is grown under long day lengths of more than 15 hours (Okumoto et al., 1996). There is some synteny for genes controlling flowering among wheat, maize, and rice, yet some of these genes appear to have different functions (Buckler et al., 2009). Heading date is also an important trait in barley. Barley floral development is controlled by the interactions between environment, genetic factors, vernalization requirement, photoperiod response and earliness genes (Takahashi and Yasuda, 1971). Wheat vegetative growth and seed yield are very sensitive to high temperature. Thus, wheat is cultivated in areas and/or seasons with moderate temperature such as temperate regions and in semitemperate areas in the world (Kumar et al., 2012). Wheat can be either photoperiod sensitive or photoperiod insensitive. Following vernalization, winter wheat plants require exposure to long day lengths (>10 hour light) to initiate flowering. Spring wheat has a wider adaptability to diverse environmental conditions because of variation in crop season length. The time to flowering of spring wheat is relatively short compared to winter wheat which generally requires an extended dormancy period during winter to complete its life cycle. It is known that vernalization insensitivity is responsible for early flowering in spring wheat (Pugsley, 1971). Winter wheat in the United States is generally sown between September and November. The heading stage occurs between the following May and June and harvesting is between June and July. Heading date in winter wheat is not greatly influenced by the date of sowing as the wheat has to be exposed to low temperatures before it breaks dormancy in the spring (Hu et al., 2005); thus spring conditions influence HD much more than autumn conditions. Spring wheat can be planted anytime that allows the crop to mature in a frost-free season. In the northern US and Canada spring wheat is planted in April or May and harvested in August providing about a 120 day growing season. 5

16 Heading (Feekes stage 10.5) in both spring and winter wheat occurs when the tip of the spike (head) can be seen emerging from the flag leaf sheath in 50% of the plants and the stage continues until the spike is completely emerged but anthesis has not begun. Anthesis follows heading and this is the stage when pollination and fertilization occur. In other cereal plants flowering and pollination may occur either before heading or after the heading stage. Heading date in wheat is controlled by vernalization, photoperiod and earliness per se (EPS) genes. Vernalization and photoperiod sensitive loci influence flowering and time to maturity (Iqbal et. al., 2007). Earliness per se genes may affect flowering time independently of vernalization and photoperiod requirements. They play a substantial role in determining the specific time during which plants flower. Earliness per se genes controls the degree of floral primodial initiation (Worland, 1996). The combination of these genes can cause wheat plants to have different flowering times and maturity, depending on the environmental conditions (van Beem et al., 2005). Vernalization response in wheat is controlled by multiple gene series designated Vrn-1, Vrn-2, Vrn-3 and Vrn-4. There are three Vrn-1 genes located on chromosome 5A (Vrn-A1), 5B (Vrn-B1) and 5D (Vrn-D1) (Sourdille et al., 2000). The three Vrn-3 loci have been located on chromosome 7A, 7B, and 7D (Yan et al., 2006; Bonnin et al., 2008) while one Vrn-4 locus has been located on chromosome 5D (Kato et al., 2003). The Vrn-2 series genes are dominant for winter growth habit wheat while Vrn-1, Vrn-3, and Vrn- 4 are dominant for spring growth habit (Yan et al., 2004). Photoperiod sensitive wheat plants require long day lengths to flower. Photoperiod response in plants has been modified as a result of crop domestication and these changes have an important role in adaptation to different environments. There are three major loci for photoperiod sensitivity in wheat located on chromosomes 2A (Ppd-A1), 2B (Ppd-B1) and 2D (Ppd-D1). The recessive alleles at these loci confer photoperiod sensitivity while the dominant allele reduces or eliminate photoperiod response (Whitechurch and Slafer, 2001). The Ppd-D1 locus is considered to have the greatest effect on photoperiod sensitivity (Kamran et al., 2014). A fourth locus (Ppd-B2) has been mapped on chromosome 7B. Low sensitivity to photoperiod is of importance in latitudes below 45 where day length can be short. Photoperiod insensitive wheat is also important in areas where spring wheat is produced under short daylengths in an effort to harvest more than one crop per year (Mohler et al., 2004). 6

17 Earliness per se genes are believed to influence wheat flowering and maturity independently of photoperiod and vernalization genes (Lewis et al., 2008). These genes are important for wide adaptation of wheat to various environmental conditions. Earliness per se genes control flowering once vernlization and photoperiod requirements are satisfied. Worland (1996) summarized the regions of the wheat genome that have EPS genes, highlighting regions on chromosomes 3A, 2B, 4D, 6B, 6D, 7B and 7B. Thus, crop variability in flowering time can be exploited and used to maximize yield potential in various environmental conditions (Lewis et al., 2008) 7

18 3. 0 Association Analysis and Genomic Selection 3.1 Association analysis (AA) The phenotype of a quantitative trait is the collective result of environmental effects, single or numerous genes (polygenes) of small effect that may interact and genotype by environmental interaction. Quantitative trait loci (QTL) analysis is a statistical procedure that utilizes phenotypic data as well as marker information (genotypic data) and or genetic maps to associate specific regions of the genome with phenotypic traits. Taken to logical conclusion, QTL analysis allows us to treat a quantitative trait as if it were a Mendelian locus and may in identification and characterization of the underlying genes. The idea of QTL analysis is not new, though in recent years the innovations of DNA markers and biometric methods for QTL analysis has led to significant advancement in QTL mapping (Asins, 2002). Association analysis (AA) is a QTL mapping technique that relies on linkage disequilibrium (LD) to determine the relationship between phenotypic and genotypic variation in population of complex origin (Breseghello and Sorrells, 2006). Linkage disequilibrium is a non-random association between the alleles of two loci such that certain combinations of alleles are more likely to occur together than other combination of alleles. Linkage disequilibrium can be observed between two loci if they are linked to each other on the same chromosome. Linkage disequilibrium can also occur when genes are not linked on the same chromosome, but statistically associated due to selection. Association analysis can be seen as an effective approach for bridging the gap between MAS and QTL analysis (Breseghello and Sorrells, 2006). Association analysis is an approach to genetic mapping amenable to complex or multi-parent populations such as those used in by plant breeders. A study by Sourdille et al. (2000) found important chromosome regions associated with flowering time in wheat on chromosomes 2B, 5A and 7B. Additionally, a study conducted by Hoogendoorn (1985) found chromosome regions controlling flowering in wheat, and identified EPS genes on chromosomes 3A, 4A, 4D, 6B and 7B. Hanocq et al., (2004) identified flowering or heading time QTLs on chromosomes 2B, 8

19 2D, 5B and 7A. Also, chromosomes groups 2, 5 and to a small extent group 7 chromosomes are generally known to have major effect on wheat development (Law and Worland, 1997). 3.2 Genomic selection (GS) Improvement of wheat quantitative traits has proved to be difficult through MAS. Genomic selection differs from association analysis (mapping) as it does not involve mapping the effect of individual genes or QTLs. Genomic selection focuses on determining the genomic estimated breeding value (GEBV) of an individual from markers covering the whole genome. A recent advance in cereal genomics research offers the opportunity for predicting phenotypes from genotypes (Varshney et al., 2006). Genomic selection uses a random effects approach to simultaneously estimate the effect of all the markers without significance testing to capture QTLs that small proportion of genetic variation yet are excluded by conventional MAS (Heffner et al., 2011). A major challenge to GS was that the number of molecular markers required to model the GEBV is larger than the number of phenotypes available for estimating their effects. This issue could be solved by using markers as random effects and mixed models to predict the effect associated with each marker (Piyasatian et al., 2007). Previously, the large number of markers required for genotyping and high costs were also limitations for implementation of genomic selection. High throughput technologies have made genotyping more practical and more affordable. Genomic selection uses LD between QTL and markers covering the genome to estimate breeding values (Habier et al., 2009). The implementation of GS is now becoming possible in many breeding programs due to decreasing costs of genotyping. The access to genotyping platforms and sequencing technologies are the foundation for execution of GS. Genomic selection makes selection possible without phenotyping by selecting superior individuals using the GEBVs. Genomic selection steps involve the use of a training population which has both the genotypic and phenotypic data, and a second population set, the selection candidates, for which only genotypic data are available. The training population allows the development of a model, which is used for calculation of breeding values (Lorenz et al., 2011). Molecular marker effects are initially estimated using the training data set with marker genotypes and the related trait phenotype of the quantitative traits (Zhong 9

20 et al., 2009). This process is referred to as training the GS model. The model can then be used to predict the value of other individuals that have been scored with the same markers. Through simulation studies it has been shown that GS has the potential to increase gain per year as compared to phenotypic selection by permitting estimation of GEBVs on selection candidates exclusive of obtaining the phenotypic data from individuals themselves (Habier et al., 2009). Phenotypic selection in plant breeding is becoming more expensive and sometimes is time consuming. 10

21 4. 0 Materials and Methods The study involved two wheat association analysis (AA) mapping populations. One was a soft winter wheat population (SWW) and the other a hard spring wheat population (HSW). 4.1 Soft Red Winter Wheat The SWW population contained two hundred eighty elite winter wheat genotypes generated by the Triticeae Coordinated Agricultural Project (TCAP) (Supplemental Table 1). The lines and one check variety, Branson, were planted in 16 different environments in the United States in 2012 and 2013 (Table 1). At each location and for each replication, the study was planted as an augmented experimental design with five blocks each consisting of 64 plots in an 8x8 grid. Each block had 56 lines and eight check plots with one plot in each row and each column. There was one replication at each site except Wooster Ohio and Warsaw Virginia where we used three replications and the study was conducted as split-plot with nitrogen rate as the whole plot and the lines as the sub-plot. All locations received 28kg of N per hectare at fall planting time as standard nitrogen rate. Two replications in Wooster and Warsaw received low nitrogen treatment of (45 kg of N per hectare) in the spring while the third replication received full nitrogen treatment (101 kg N per hectare). The other locations received full nitrogen in the spring. Heading date (HD), Feekes growth stage 10.5, data was collected from each environment in 2012 and Heading date, as Julian days, was recorded when the head (spike) on 50% of the wheat plants had completely emerged from the flag leaf. Prior to any further data analyses the data from each plot in each block within each replication (set of 320 plots) was adjusted for block effect based on an augmented design: Ý=Y ij - (x i- x ) 11

22 Where Ý is the adjusted trait value of j th genotype in the i th block, Y ij is the raw data, x i is the mean of check plots in the i th block; x is the mean of the checks over all blocks within a replication. The block adjustments were done using SAS software (SAS Institute 2011). The correlation of line means between environments was obtained from PROC CORR. Because just four of 28 total environments (HSW and SWW trials) had two replications and 24 had one rep we decided to average over reps prior to all ANOVAs. PROC MIXED was run using the adjusted means for each environment to get best unbiased linear predictions (BLUPs) of the main effects of genotypes, environments, genotype by environment interaction (GEI) effects and estimates of variance components. The model was: y ij = μ + g i + e j + ge ij where: y ij is the corresponding phenotype of the i th genotype in j th environment, μ is the grand mean, g i is the main effect of i th genotype, e j is the main effect of j th environment, ge ij is the GEI effect which is confounded with error effects. 4.2 Hard Spring Wheat Population The hard spring wheat population contained 256 elite spring wheat photoperiod sensitive genotypes generated by the TCAP (Supplemental Table 2). For this analysis we used 249 lines that had both phenotypic and genotypic data. The wheat genotypes were planted in twelve different environments in the United States of America in 2012 and 2013 and in Arusha Tanzania in 2013 and 2014 (Table 2). The study was planted in an augmented-rcb design with one replication at each location. Heading date was recorded at the date of Feekes stage 10.5 as days from planting. Phenotypes were adjusted for block effects and BLUPs were obtained as described for the SWW population. 4.3 Heritability and GEI analysis We used means over replications to estimate heritability of HD and was estimated as: 2 g H 2 2 g e 12 error

23 where H is heritability, σ 2 g is variation due to genotype, σ 2 error is error variance and is due to both GEI and error variance. The BLUPS of the GEI effects were placed in a matrix and we used Wards minimum variance as the clustering criteria to cluster the environments using PROC CLUSTER. An Additive Main Effects and Multiplicative Interaction (AMMI) analysis was conducted using SAS and biplots were created using scores for the first two principal components (Zobel et al., 1988). 4.4 Genotyping For SNP analysis a novel 90k Infinium chip (90k iselect) was used to genotype both populations. For the SWW we used data from 13,228SNPs that had call rates greater than 70% and had putative map positions. For the HSW population we used 20,189 markers that had call rates greater than 70% and putative map positions. 4.5 Association Analyses and Genomic Selection Association analysis (AA) was performed using Genomic Association and Prediction Integrated Tools (GAPIT) a package created in R the environment (Lipka et al., 2012). A mixed Linear Model (MLM) that includes both fixed and random effects was used to perform AA. We used values from a principal components analysis generated by GAPIT to correct for population structure, and a kinship matrix(k) to account for relatedness between individuals. The kinship matrix was generated automatically using VanRaden (2008) method in this analysis. The AA model used by GAPIT was: Y = Xβ +Qw + Sα +Zv + e Where Y is a vector of observed phenotypes, Xβ is non-genetic fixed effect (mean); Qw is a fixed effect with Q a principal component scores, and w a vector of principal component effects, Sα is a fixed effect with S a matrix of marker scores, and α a vector of marker effects; Zv is random effect with Z a matrix relating observations to their polygene effect, and v a vector of polygene effects. 13

24 In running the analysis the genotypic data for both spring and winter wheat data were in the hapmap format. The phenotypic data were the BLUPS from HD phenotypes collected over environments. Five principal components were used in the model, and the regular linear mixed model setting was used for this analysis. Each individual was considered as a group and was performed by setting the number of groups equal to total number of individuals (group.from=n and group.to=n). The total number of markers used for association analysis was 13,228 for the SWW and 20,189 for the HRW populations. For the SWW we analyzed HD i) over all environments, ii) just the northern environments (North), and iii) just the southern environments (South). For the SWW we analyzed HD i) over all environments, ii) just the Californian Tanzania environments (CATA), and iii) just the other North American (NA) environments. In the SWW population 77 markers were found to be highly significant (p<0.0005) for at least one trait and were selected for LD analysis to assess their independence. In the HSW population 122 markers were highly significant (p=0.0005) for at least one trait and were selected for LD analysis. Linkage disequilibrium analysis between pairs of SNPS was calculated using the R packages genetics (Warnes and Leisch, 2006) and LDheatmap (Shin et al., 2006). Genomic prediction was performed using the R package rrblup (Endelman, 2011). To obtain GEBVs 11,926 markers that were scored in both winter and spring wheat was used. The markers were imputed for missing values using the A.mat function in rrblup. Genomic selection was performed using the R package rrblup (Endelman, 2011). The rrblup package uses the Mixed. Solve function (Endelman, 2011) to obtain GEBVs. We used ten-fold cross validation to obtain GEBV whereby the 249 genotypes (HSW) and 275 genotypes (SWW) were subdivided randomly into ten disjoint subsets. Nine subsets were used to build the GS model and that model was used to predict the value of the remaining validation subset: the predicted values were then correlated with the observed phenotype. 14

25 5.0 Results 5.1 SWW phenotypes and GEI We analyzed HD expressed in Julian days in the SWW population. Winter wheat in southern environments breaks dormancy before wheat in the northern environments and thus has an earlier HD when expressed as Julian days. This does not mean that the period between breaking dormancy and heading is shorter in the southern environments than in the North. Heading date expressed in Julian days does not estimate the total vegetative period of an observation. The use of Julian days in the SWW population will have a large impact on estimates of the main effect of environments, but little effect on the relative value of the main effect of genotypes, QTL or GEI. In the analysis of variance of HD in the SWW population over all environments we found both genotype and environment effects to be highly significant (p < ) (Table 3). As we average over replications prior to the ANOVA the error variance is due to GEI and error effects: genetic variance was 1.58 times greater than the GEI (error) variance. The results show the existence of genetic variation for the trait and that genotypes responded differently when exposed to different growing environments. We used clustering and AMMI analysis of the GEI matrix to place the 16 SWW environments into clusters where GEI would be maximized between clusters and minimized within a cluster. This produced two distinct clusters (Fig. 1, Table 1). One cluster contained the six Ohio environments plus Missouri 2013 and was termed the North cluster. The other cluster had all the environments from the southern US and was termed the South cluster. Environments that differed only by nitrogen treatment always clustered together. The bi-plot from the AMMI analysis (Fig. 2) shows a similar pattern. In the South cluster, all of the environments had negative main effect except 13KYM while the environments in the North cluster all had positive main effects (Table 1). Using an ANOVA, the GEI variance was partitioned and 58% of the 15

26 total GEI variance was due to genotype x cluster effects and 42% of the GEI variance was due to genotype by environment effects within clusters. The estimated heritability for HD in the SWW population was 0.96 over all 16 environments, 0.93 in the North cluster, and 0.96 in the South cluster (Table 3). More genetic variance for HD was observed in the South than in the North (Table 3). The ranges of genotype main effects for HD were 13.3, 17.8, and 8.3 days over all environments, in the South, and in the North, respectively (Supplementary Table 1). This corresponds to the greater genetic variance in the South than the North. There is a significant positive correlation (r=0.83) between the genotype main effects over the North and over the South environments. There was a positive correlation (r=0.85) between the main effect of genotypes and their GEI effects with the South environment (Figs. 2, 3): late genotypes had positive interactions with south environments while early genotypes had negative GEI with south environments. There was a negative correlation (r=-0.80) between the main effect of genotypes and their GEI with the North environments (Figs. 2, 4): late genotypes had negative GEI with North environments, while early genotypes had positive GEI with North environments. These results indicate that early heading genotypes are later than expected based on main effects when planted in the North, and they are earlier than expected when planted in the South. The late genotypes are earlier than expected based on main effects when planted in the North and they are later than expected when planted in the South. 5.2 HSW phenotypes and GEI For the HSW population HD was recorded as days from planting to Feekes growth stage This system has a different interpretation than when HD was recorded as Julian days in the SWW population. Here HD is an estimate of the vegetative period of a wheat line. The analysis of variance over all environments showed that there were significant genotype and environment effects (Table 4). The GEI variance (error) was 1.78 times greater than the genetic variance (Table 3) indicating the GEI was more prevalent in the HSW population than in the SWW population. The results show the existence of genetic variation for the trait and that genotypes responded differently when exposed to different growing environments. 16

27 Through cluster and AMMI analysis we placed the 12 HSW environments into two distinct clusters (Fig. 5, Table 2). One cluster contained eight environments; seven from the United States and one environment from Canada and this cluster was named the North America cluster. The second cluster contained two environments from 2012 Davis, California, and two environments from Arusha, Tanzania and was named the CATA cluster (California-Tanzania). The bi-plot from the AMMI analysis (Fig. 6) also shows the same pattern of environment clustering. The four environments in the CATA cluster have very different main effects with the Tanzania environments having large negative main effects and the Davis, California sites having large positive effects (Table 2). The environments in the North American cluster have similar main effects whose absolute value is small compared to those of the CATA environments. Using an ANOVA, the GEI variance was partitioned and 86% of the total GEI variance was due to genotype x cluster effects and 14% of the GEI variance was due to genotype by environment effects within clusters. Genetic variance for HD was much greater in the CATA environments than in the NA environments (Table 4). The later heading HSW genotypes tend to have a positive interaction with the CATA environments and a negative interaction with the North American environments, while earlier genotypes tended to have the opposite GEI pattern (Figs. 7 and 8). The estimated heritability for HD was 0.88 over all environments, 0.96 in the CATA cluster, and 0.91 in the North America cluster (Table 4). The range of genotype main effects was 17.9, 40.4 and 8.8 days over all environments, in the CATA and North American clusters, respectively (Supplementary Table 2). This corresponds to the greater genetic variance in the CATA than the North American cluster (Table 4). 5.3 Association analyses For the SWW and HSW population we analyzed 13,228 and 20,189 markers respectively (Tables 5 and 6). In the SWW populations approximately 38%, 52%, and 10% of the markers were from the A, B, and D genome, respectively and in the HSW population approximately 39%, 51% and 10% of the markers were from the A, B, and D genome, respectively. The correlation of markers per chromosome between the two populations was 0.98 indicating similar genome coverage in each population. 17

28 We considered both the false discovery rate (FDR) probability values and the unadjusted probability value from GAPIT when assessing QTL and genetic structure of the traits. The FDR is very conservative and over all traits and both populations just one marker had a FDR probability less than The heritability for all traits was high (>0.87) so it seems almost certain that many markers with FDR probability value greater than 0.05 are tagging real QTL whose effects are simply not large enough to pass the very stringent FDR criteria. One of our objectives was to compare genetic architecture over environments and between winter and spring wheat and FDR seemed poorly suited for that objective due to a seemingly high type II error rate. Thus we used unadjusted probability values to identify significant (p < 0.05), very significant (p < 0.005) and highly significant (p < ) SNP-QTL association. For the SWW population we analyzed three HD traits: HD over all environments, over just the North environments, and over just the South environments. We found a total of 1,793 markers (6.5 %) were significant (p<0.05) for at least one of the three HD traits and 348 were significant for all three traits (Table 6). The correlation of allele effects for the 857 markers that were significant (p<0.05) in either the North or in the South clusters of environments was Regressing the allele effect of these markers from the South on the allele effects from the North produced a significant (p<0.05) regression equation of y= x, suggesting allele effects in the South are 2.14 times greater than in the North. For the HSW population we also analyzed three traits: HD over all environments, over just the North American environments, and over just the CATA environments. We found a total of 1,646 markers (8.2%) were significant (p<0.05) for at least one trait though only 295 were significant for all three traits (Table 6). The correlation of allele effects for the 1,345 markers significant (p<0.05) in either the North America or in the CATA environments was Regressing the allele effect of these markers from the CATA environments on the allele effects of the North American environments produced a significant (p<0.05) regression equation of y= x, suggesting that allele effects in CATA were 5.11 times greater than in the North American cluster. In both populations the distribution of the significant (p<0.05) markers over chromosomes was very similar to the distribution of all markers over chromosomes: 30-45% from the A genome, 47-55% from the B genome, and 8-17% from the D genome. In the SWW population 77 markers were very significant (p<0.0005) for at least one trait and had 18

29 a minor allele frequency (MAF) value greater than These markers were selected for LD analysis to identify independent QTLs. The identification of independent QTLs within a chromosome was straightforward as the r 2 values between markers were either greater than 0.4 or were close to zero. The consensus map position and the LD analysis suggested there were seven very significant (p<0.0005) and independent QTL for HD in the SWW population (Table 7). Five of these seven QTL were very significant (p<0.0005) for all three HD traits while two (QTLs 1 and 2, Table 7) were very significant overall environments and just in the South cluster of environments. These seven very significant QTLs had fairly low r 2 values and allele effects (Table 7). QTLs 1 and 2 are represented by markers that appear close together in the consensus map but were in low LD in the SWW population. In the HSW population 119 markers were very significant (p<0.0005) for at least one of the three traits and had MAF greater than These markers were selected for LD analysis to identify independent QTLs. As in the SWW population, the HSW LD r 2 for markers from the same chromosome were either greater than 0.4 or close to zero. The consensus map position and the LD analysis suggested there were seven very significant (p<0.0005) and independent QTL for HD in the HSW population (Table 8): the marker associated with QTL 2 was also significant using a FDR of There were 11,926 markers scored and analyzed in both the SWW and HSW populations. Of these 532 were significant (p<0.05) for at least one of the three traits in the SWW population, 647 were significant for at least one of the three traits in the HSW population, and 713 were significant for one of the three traits in the SWW or the HSW population. The correlation of allele effects for these 713 markers between the three SWW and the three HSW traits ranged from to Only two genomic regions showed strong evidence of being very significant (p<0.005) for HD in both populations, even though 13 markers were very significant (p<0.0005) in just the SWW population and 12 markers were very significant in just the HSW population. One coincident QTL was identified by the same marker in both populations and was located on chromosome 4A, though it did not have a probability value < for any trait in either population. The second coincident QTL was on chromosome 5A and appeared to correspond to QTL 6 from the HSW, and QTL 5 from the SWW populations. The best markers for each of these QTL are close together on the consensus map, though neither was scored in both populations. One marker from this 19

30 region was scored in both populations and had a probability value of for HD in the SWW population and for HD in the HSW population. This common marker was in LD with the QTL markers in the SWW and HSW populations. Four other regions from chromosomes 2B, 3B, 4D, and 5B were significant (p<0.05) in one population and had probability values between and in the other population. The 2B region appears to correspond to QTL 1 in the HSW population. The seven very significant QTL in the SWW population and the seven QTL from the HSW population all had r 2 estimate of allele variance that were values less than 0.13 and the absolute value for allele effects did not exceed 4.69 days (Table 6). The r 2 values and allele effects were higher in the SWW and HSW clusters of environments that produced the greatest range of HD and greatest genetic variance (e.g. the South SWW cluster and the CATA HSW cluster) 5.4 Genomic selection Genomic selection models were built for each trait in each population. The GEBVs were obtained using 10 fold sampling and cross-validation. These GEBVs were then correlated to the BLUPs of the phenotypes (Table 8). The correlation of GEBVs and BLUPs ranged from 0.46 to 0.66 and the relative efficiency of GS ranged from 0.48 to These values were greater in the South cluster of environments of the SWW population than in the North cluster, and in the CATA cluster of environments of the HSW than in the North America cluster. 20

31 6.0 Discussion The process of flowering in wheat plants is controlled by vernalization, photoperiod response and earliness per se genes. In this study vernalization was confounded between the winter (SWW) and spring (HSW) populations. Both populations contained photoperiod sensitive lines. The results of analysis of variance over environments showed that there were significant genotype, environment, and genotype x environment interaction for HD in both populations. The heritability of HD was high in both populations, within each cluster of environments, and within each population (Tables 3 and 4). These estimates of heritability are consistent with reports of high heritability for HD (0.75) in the literature (Eid, 2009).These results indicate that selection for early flowering based on phenotypes would be expected to result in genetic improvement for the trait. Selection for early flowering in spring wheat is very important to ensure high grain production and quality (Iqbal et al., 2007). Clustering placed the environments from each population into distinct groups. In each population, one cluster of environments produced a larger range of HDs (South cluster for SWW and the CATA cluster for HSW) while the other produced a significantly smaller range of HDs. However, in both populations the phenotypic values were highly correlated between the two clusters. The separation of environments into clusters based on GEI appears to be due to a difference in genetic variation within specific environments for HD between the clusters rather than lack of correlation of phenotypes. The differences in variation for HD between the South and North clusters of environment for the SWW population is possibly because of the difference in magnitude of the effects of photoperiod (Ppd) and earliness per se (EPS) genes between the two clusters. Ppd and EPS genes cause variation for HD but their effects seem to vary between the two clusters. In the SWW South cluster the Ppd genes that do segregate in the SWW population are likely more influential than in the North as day lengths are shorter in the South 21

32 than the North when the SWW breaks dormancy. Kamran et al. (2014) reported that the Ppd genes influence heading more in lower latitudes than Northern latitudes due to the shorter day lengths in the South. The greater effect of these Ppd genes may in part cause greater range of HD in the South than the North. In the north the SWW lines break dormancy later in the calendar year than in the South and under longer days so the day length requirement for the Ppd genes is met sooner and thus the Ppd genes cause less variation for HD than in the South. EPS genes would likely have similar effects in the North and South. The differences in variation shown by the two clusters of environments for the HSW population could also be due to difference in the effect of Ppd genes and the effect of growing temperature between the two clusters. The Ppd genes effect changes due to the presence of precise photoperiod response genes and the latitude of the growing environment (Kamran et al. 2014). The environments in the North American cluster of the HSW population all experience long days immediately after crop emergence in early May such that day lengths quickly meet the flowering requirement and Ppd genes likely have a small effect on heading. The California environments in the CATA cluster were planted in October under short days and Tanzania environments always experienced short days regardless of planting date. Thus for photoperiod sensitive wheat cultivars there will be a delay in flowering in these short day environments. In short day environments the Ppd genes affected HD more than in the long day environments. Thus the effect of day length on Ppd genes may have caused greater variation for HD in the CATA environments as some lines needed longer day lengths to flower than others due to their Ppd genotypes. This result is supported by the previous findings. Worland and Snape (2001) suggested that adaptation of spring wheat to different agro-climate conditions is highly influenced by Ppd genes. It has also been reported that lack of fulfillment of photoperiod requirement delays flowering in photoperiod sensitive spring wheat (Kamran et al.2014) and this delay would have occurred in the CATA environments. Photoperiod response genes influence the transition of wheat plants from vegetative phase to reproductive stage by affecting phonological processes, spikelet initiation and stem elongation (Kirby, 1988; Miralles et al., 1998). 22