Genetic dissection of complex endosperm traits

Review

Genetic dissection of complex endosperm traits Xuefeng Wang1,2, Chenwu Xu1, Rongling Wu3 and Brian A. Larkins4 1

Jiangsu Provincial Key Laboratory of Crop Genetics and Physiology; Key Laboratory of Plant Functional Genomics of the Ministry of Education, Yangzhou University, Yangzhou 225009, China 2 Department of Epidemiology and Biostatistics, Case Western Reserve University, Cleveland, OH 44106, USA 3 Center for Statistical Genetics, Departments of Public Health Sciences and Statistics, Pennsylvania State University, Hershey, PA 17033, USA 4 Department of Plant Sciences, Forbes Building, Box 210036, University of Arizona, Tucson, AZ 85721, USA

The endosperm of plants is a major source of food, feed and industrial raw materials. The genetic analysis of endosperm traits poses numerous challenges due to the endosperm’s complex genetic composition and unique physical and developmental properties. Modern molecular techniques and statistical methods have greatly improved the mapping of quantitative trait loci underlying endosperm traits and have led to revolutionary insights regarding epistatic and epigenetic effects. This article describes the current state of the methodologies used in the genetic dissection of endosperm traits and highlights practical issues and statistical concepts and procedures. Endosperm origin and function The endosperm of seeds is important both economically and scientifically. It is a major source of food and feed for human and animal nutrition, and it provides raw materials for manufacturing many industrial goods, including biofuels. The endosperm is a good model system for investigating the mechanisms regulating important biological processes in plants for a variety of reasons: (i) the cytological features of endosperm formation are well known; (ii) it has a determinant development with a short period of growth and maturation, making it ideal for developmental and cell biology studies; and (iii) it is a simple tissue composed of a few specialized cell types that produce large quantities of storage metabolites, making it a useful target for biochemical and molecular biology studies. The focus of this review is the genetic analysis of endosperm traits; for a more in depth consideration of endosperm development and function, see Refs [1–4]. Endosperm is a unique feature of the flowering plants (angiosperms). It functions primarily as a storage tissue for nutrients, for example carbohydrates, proteins, lipids, nucleic acids and minerals, that nourish the embryo during its development and germination. The amount of endosperm in a mature seed is variable. For example, it comprises a large portion of the seed in cereals, such as rice (Oryza sativa), maize (Zea mays) and wheat (Triticum aestivum), and it is prominent in the seeds of some dicots, such as caster bean (Ricinus communis). In other species, Corresponding authors: Xu, C. ([email protected]); Wu, R. ([email protected]).

for example Arabidopsis (Arabidopsis thaliana), the endosperm is almost completely absent in the mature seed. However, whether the endosperm persists or is ephemeral, it is vital for embryo development in much the same way the placenta is essential for mammalian embryos. Endosperm is formed during a double fertilization event involving two sperm cells from the male gametophyte (pollen grain) and the haploid egg cell and homodiploid central cell from the female gametophyte (embryo sac). The homo-diploid central cell gives rise to the endosperm; consequently, the embryo is diploid (2C) and the endosperm is triploid (3C). As will be explained, the triploid nature of the endosperm creates unusual gene dosage and gene interactions that affect its development. Some angiosperms form endosperm from different numbers of polar nuclei, which results in ploidies of 2C, 3C, 5C, 9C and 15C [5]. Other seed plants (gymnosperms) also form a nutritive tissue that supports embryo growth, but it arises from the haploid (1C) cells of the female gametophyte. Most angiosperms manifest the so-called ‘nuclear’ or Polygonum pattern of endosperm development [5], resulting in a 3C endosperm. This mode of development has four distinct stages, namely the syncytial, mitotic, differentiation and maturation stages [4]. After fertilization, the primary endosperm nucleus undergoes multiple rounds of division without cytokinesis, resulting in a syncytium of multinucleate cytoplasmic domains. Cellularization of the syncytial endosperm is usually complete several days after pollination and is generally followed by a period of mitosis. During the mitotic phase, endosperm cells have a standard cell cycle, and the newly formed daughter cells maintain their initial triploid DNA content. This phase determines the number of cells in the endosperm and might be important for grain yield in cereal crops [6,7]. The subsequent differentiation of the endosperm results in a small number of cell types, some of which function in the transport of sugars and amino acids into the endosperm, whereas others store metabolites, protect the embryo and prepare for germination [3,8]. Frequently, the genome of the storage cells undergoes several rounds of endoreduplication; the resulting polyploidy is thought to be associated with an enhanced capacity to synthesize and store metabolic reserves [9].

1360-1385/$ – see front matter ß 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.tplants.2009.04.004 Available online 21 June 2009

391

Review The origin of the endosperm is a long-standing question [10]. One hypothesis holds that it arose as the consequence of heterochronic changes in the development of the female gametophyte. The ‘sexualization’ of gametophytic cells through fertilization with a sperm nucleus could have created genetic or polyploidy-related benefits [10]. A second hypothesis posits that the endosperm developed from a second embryo that degenerated into a nutrient storage tissue [11]. The discovery of double fertilization events in the gymnosperms Ephedra and Gnetum, precursor seed plants more closely related to conifers than angiosperms, is considered evidence of the second embryo hypothesis [12,13], whereas recent studies showing that solely maternally derived endosperm has features of wild-type endosperm provide evidence for the gametophytic origin of endosperm [14,15]. Linkage analysis is still the first choice in genetic studies of endosperm traits The application of molecular marker techniques has shifted the focus in quantitative genetics from estimating aggregate genetic effects to localizing determinant genomic regions, or quantitative trait loci (QTLs). QTL mapping of endosperm traits, however, faces additional statistical and experimental challenges. The first issue is that standard laboratory procedures cannot determine the genotypic state of the triploid endosperm (Box 1). DNA extraction from a single endosperm of some seeds can be troublesome, and it is impossible to distinguish two heterozygous genotypes QQq and Qqq using common dominant or codominant markers. Therefore, to make an interval mapping (IM) method that searches for the QTL between two anchored markers applicable [16], it is necessary to predict the genotypes of a putative endosperm QTL based on the flanking marker information from the maternal plant. This ‘in silico’ genotype inference system has been well established using both backcross [17] and F2 designs [18] (Box 2). A two-stage design [19] was further proposed to draw markers from both maternal plants and their embryos. With more genotype combinations produced, the F2 design tends to be more informative than the backcross, and the twostage design more informative than the design based only on the maternal genotypes [20]. The imputed QTL genotypes allow phenotypes to be expressed by a linear model containing previously introduced triploid genetic parameters and a random error. Intuitively, the equation can be treated as a simple linear regression with coefficients estimated by ordinary or weighted least-square methods [18]. The same analysis is then extrapolated to all putative loci to perform a whole genome scan, where significant QTLs can be detected by the classic regression inference techniques. Alternatively, the phenotype–genotype model can be formulated as a statistical mixture model with different QTL genotypes as components of a normal mixture, in which the maximumlikelihood estimates (MLEs) of the parameters are typically computed by implementing the expectation maximization (EM) algorithm [19,21] Several caveats must be noted regarding endosperm QTL mapping. First, investigators need to ensure that the method chosen is compatible with the mode of inheritance of the endosperm trait being studied. Because seeds 392

Trends in Plant Science Vol.14 No.7

represent a new generation, the expression of an endosperm trait is impacted by two genetic systems. Some traits such as grain quality attributes are more likely to be controlled by the endosperm genome than by the maternal genome. This fact was often ignored in many earlier studies [22–24] owing to the inability of most QTL-mapping software packages to handle endosperm traits. Use of a simple diploid model might result in confounded effects and inflated sampling variance, and consequently biased estimation and loss in resolution power [20]. Second, for measurement convenience, seeds collected from a single plant are often bulked as a composite sample. However, it is incorrect to simply use the sample mean to represent the phenotype of a single endosperm in the genetic mapping because each endosperm has an independent genotype. The mean-based observation data has recently been incorporated into the mapping system based on a multinomial-based mixture sampling procedure [25], obviating the need to assess the phenotype for each seed. Finally, because multiple genetic parameters are involved, endosperm mapping is more data-hungry than traditional approaches. One could consider genotyping two generations or utilizing highly informative segregation populations to further improve performance of the QTL mapping [26–28]. There are also a variety of refinements to statistical models that will be discussed in the following sections. Searching for needles in a genome haystack: multilocus and multi-trait mapping Although IM is simple to implement, this method might have low resolving power because it ignores the loci outside the interval. The sampling error of a target QTL will be inflated if multiple QTLs are unlinked. And if they are linked, the main effect QTL tends to mask other QTLs in the population and sometimes even creates false identification, generating a ‘ghost peak’ [29]. This led to the development of composite IM as a way of increasing resolution by performing an interval search while including the nearby genetic markers as cofactors in multiple regression to absorb the background effects [30]. These markers can serve as the proxies of other QTLs, but like in any other regression problems, the choice of optimal number and combination of cofactors remain a challenge. And different cofactor selection methods often lead to different, sometimes conflicting mapping results [31]. Instead of focusing on one interval at a time, the multiple-IM (MIM) technique [32] performs a multidimensional search of several putative loci and associated epistatic effects, which allows direct estimation of the number, positions and effects of QTLs. MIM has been extended to accommodate the triploid endosperm model by its developer [20]. Although MIM was an important advance, it too makes it difficult to investigate all the loci in the genome simultaneously. In fact, an exclusive search of all possible statistical models is computationally impossible, and existing methods are only able to explore a subset of these models. One common way of subsetting is to use the ‘windows moving’ strategy that concentrates on a small genomic fragment each time. But such a process makes it hard to develop an integrative genome-wide QTL model and impossible to check pair-wise epistatic interactions

Review

Trends in Plant Science

Vol.14 No.7

Box 1. Inheritance features and quantitative genetics of endosperm traits ‘Carl Correns confirmed Mendelism while studying hereditary traits in the endosperm . . .’ [79] Although the phenomenon of double fertilization was discovered as early as 1899, the inheritance features of endosperm were not fully recognized in quantitative genetics until two decades ago [80,81]. Endosperm traits have several unique properties that differ from those of diploid traits. First, for a locus with two alleles, Q and q, the triploid endosperm has four possible genotype combinations, QQQ, QQq, Qqq and qqq, versus three, QQ, Qq and qq, for the embryo (Figure I). Consequently, endosperms from reciprocal crosses will have a different genetic makeup. Second, the endosperm represents a generation beyond the maternal plant. Thus, genotypes of the endosperms harvested from the same plant might be distinct [18,19], and there are three genotypic effects associated: a, d1 and d2, where a is the additive effect or the mean substitution effect of Q to q; d1 is the dominant effect of QQ to q, also known as the first dominant effect; and d2 is the dominant effect of Q to qq, also called the second dominant effect. The genotypic values of the four genotypes and their relationship with the genetic effects are demonstrated in Figure II. Therefore, the phenotypic value of an endosperm trait (yi) can be described by the linear model: y i ¼ m þ x 1i a þ x 2i d 1 þ x 3i d 2 þ ei

In the interval QTL analysis, the xs are missing and need to be imputed based on the flanking marker genotypes. The position of a QTL is where the fitted model is found significant, which is usually indicated by log-likelihood ratio (LR) profiles.

[Equation I]

where m is the overall mean, e is the residual error and the xs are the indicator variables modeling the genotypes based on the scales in Figure II. For example, for genotype QQQ: x 1i ¼

3 and x 2i ¼ x 3i ¼ 0 2

[Equation II]

Figure I. Genetic constitution of seeds with triploid endosperm.

Figure II. Genotypic values of endosperm QTL genotypes (triploid model).

outside the tested ‘window’. Another strategy is the twostage process, where the first stage is used to create a shortlist of promising loci using a quick single-locus search or other one-dimensional methods, followed by a second stage where more elaborate methods can be used [31]. Recently, considerable attention has been given to sparse

regression techniques, such as LASSO [33,34], which greatly facilitate the optimum global search for multiple genetic loci. By shrinking non-significant coefficients to zero, LASSO is able to do parameter estimation and variable selection simultaneously. The newly proposed refinements, such as ‘fused LASSO’ and ‘group LASSO’, provide 393

Review

Trends in Plant Science Vol.14 No.7

Box 2. Experimental designs and genetic compositions An initial step in any QTL analysis is to choose a suitable mating design that can extract the most information subject to cost or experimental constraints. Most mapping designs start with the development of F1 hybrids (Qq) that are derived from crosses between two homologous lines, P1 (QQ) and P2 (qq), usually with extreme phenotypes for the trait of interest. F1s can be selfed (denoted by  in Figure I) further to create F2 (Ft) families or crossed with one (or both) of the parents to form backcrosses. Because the genotypic characterization of a triploid tissue is difficult to confirm, endosperm is generally not genotyped. For example, in an F2 design, each F2 is genotyped and the seeds harvested from the F2 are scored for phenotypes. Because the endosperms belong to F3 progeny, this is also known as an F2:3 mapping population. Consequently, one might consider joint maternal–endosperm mapping based on the six possible genotypic combinations, that is QQ–QQQ, Qq–QQQ, Qq–QQq, Qq–Qqq, Qq–qqq and qq–qqq [82]. Marker

information can also be derived from the maternal plants and the derived embryos, establishing a two-stage hierarchical design [19]. In terms of inferring the conditional probabilities of endosperm genotypes, the two-stage design is clearly more informative than the design that only utilizes the maternal marker genotypes. To fully dissect the endosperm genotypic effects, it is often necessary to combine information from multiple crosses in one analysis. As illustrated in Figure I, an F2 can be backcrossed to both parental lines (but with pollen from P1 and P2), as in the classic North Carolina III (NCIII) mating scheme, or further crossed to the F1, as in the triple testcross (TTC) mating scheme, to create extra populations with distinct segregating components. Such schemes are expected to be more powerful but involve laborious crossing work. The choice of which design to use is based on the cost versus information tradeoff, as well as the goals of the experimenter.

Figure I. Genetic compositions of common experimental designs.

promising solutions for handling high correlation and cluster structure among dense single-nucleotide polymorphism markers [35,36]. Genetic improvement of endosperm traits usually involves several different loci in a selection index. These traits are often correlated owing to shared genetic or environmental factors. Joint use of multiple traits in an analysis was shown to provide more precise estimates and more powerful QTL detection than trait-by-trait analysis [37–39]. Multi-trait analysis is particularly useful in endosperm QTL mapping because multiple correlated traits provide additional information regarding each other and thus reduce the effect of error variance without increasing the population size. In addition, it allows testing of close linkage versus pleiotropy and modeling of the causal relationships among different traits and QTLs [40]. However, the above merits can be compromised by including too many uncorrelated traits, where the information gain is insufficient to compensate for the increased degrees of freedom. Also, it should be noted that most attempts to deal with multiple traits were based intuitively on the multivariate regression model, which assumes the same genetic model for all traits. One less common technique, Zellner’s ‘seemingly unrelated regression’ (SUR), has been found to be more appropriate for nonpleiotropic QTLs because it allows the separate fit of each trait but still uses a joint correlation structure [41]. 394

Time matters: endosperm development and kinetics The life of the cereal endosperm is determinate and short, but the developmental processes are complex. A large number of mutants, such as crinkly4 [42] and arabidopsis crinkly4 (acr4) [43], have been isolated in maize and Arabidopsis, each providing a point of entry into the genetic network of endosperm development. Although mutant analysis has been fruitful, the temporal characteristics of the gene networks and the mechanisms controlling the developmental patterning remain poorly understood. A thorough and more integrated analytical approach is needed to create an accurate portrait of the dynamic genetic architecture regulating endosperm development. Most developmental traits are measured longitudinally over different time points. Although the above-mentioned multi-trait mapping methods can be applied to account for the multivariate nature of the phenotypic outcomes, a major disadvantage is that the numbers of model parameters increase explosively with the growing dimensions. This can be computationally intractable with even five time points. Further, the results are not biologically meaningful because the temporal effects and the biological principals are ignored. Finally, the ordinary multi-trait approach does not lend itself very well to handling autocorrelation between successive time points. A new paradigm for bridging these gaps, known as ‘functional mapping’, has been proposed [44–46]. By embedding kinetic models underlying

Review the developmental processes into traditional mapping techniques, functional mapping can estimate the dynamic patterns of genetic effects and fully dissect the relationships between genes/genotypes and development. It can be conceptually visualized by using characteristic trajectories or profiles to represent the phenotypic responses. Via parsimonious modeling of temporal trends and the covariance structure (using a structured covariance matrix), computation becomes feasible and the results are more stable. It should be noted that the appropriate mathematical equations must be chosen based on the specific developmental process, and nonparametric approaches, such as random regressions, might be considered when there is no obvious functional form. For example, two models were formulated to quantify the processes involved in maize endosperm development, where the logistic growth curve and a set of differential equations were used to describe the kinetics of mitosis and DNA endoreduplication, respectively [47]. Within the framework of functional mapping, a series of hypothesis tests can be further formulated to assess the control of a QTL over the entire growth trajectory or at specific time intervals and to understand the interplay among genes, developmental expression and environments. A tale of three systems: dissecting maternal effects in seeds The development of endosperm is controlled by unique zygotic genetic factors. Because the endosperm, along with the embryo, is enclosed in the maternal reproductive organs, mechanisms must have evolved to coordinate endosperm and embryo development with physiological and biochemical changes in the maternally controlled environment [48]. Maternal effects and their interactions with zygotic effects should constitute an important part of these mechanisms. An increasing number of mutations that cause maternal effects have been isolated over the past few years [49–51], suggesting that there might be a larger diversity of maternal effects than previously thought [52]. Maternal effects are often manifested in differences in the size and shape of the endosperm [53]. Using an analysis of different generations derived from inbred lines, significant evidence for maternal control of endoreduplication in maize endosperm has been detected [54]. Further molecular evidence confirmed that the maternal expression of two genes is required for endosperm development [55]. Large maternal effects were also reported for seed quality traits [56,57]. It should be noted that although the endosperm acts as a sink for nutrients that nourish the embryo, it is inaccurate to consider endospermic influence as one class of maternal effect, simply because the endosperm represents a new generation. The maternal influence on endosperm development could be exerted through two routes. First, in double fertilization, the central cell as progenitor of the endosperm contributes substantial cytoplasm to the endosperm. Thus, maternally derived mRNA and proteins that are located at specific subdomains in the central cell could influence the expression of endosperm traits [58]. Second, the maternal organs provide the nutrients, hormonal signals and environment necessary for endosperm development. Although cell proliferation of the integument is largely independent from endosperm

Trends in Plant Science

Vol.14 No.7

growth, the prevention of cell elongation in the integument by the transparent testa glabra2 mutation is found to inhibit endosperm and seed growth [48]. Integration of information about maternal contribution is advantageous in traditional quantitative genetics owing to the increase in the portion of phenotypic variation that can be explained. Many early studies of maternal effects relied exclusively on diallel analysis of reciprocal crosses and failed to isolate effects arising from other phenomena [59]. Maternal contribution has been assessed in genetic models for quantitative seed characters [60] and triploid endospermic traits [61] by crudely estimating basic variance components, including maternal genetic effects, cytoplasm and the direct embryo/endosperm genetic effect. Incorporation of both direct and maternal (indirect) effects in a QTL analysis is particularly meaningful for endosperm traits to avoid confounding effects from distinct, albeit correlated, generations. The option to ‘opt-out’ of maternal effects has recently been suggested [27]. With two well-devised designs, the authors showed the possibility of estimating the direct effects of endosperm QTLs with or without minimal maternal influence. Another group introduced a joint QTL genotype system into the standard mixturemodel interval scan [62,63], so that each combination of maternal and nested endosperm genotypes represented a mixture component and their frequencies were derived from maternal markers. The results enable one to ascertain the actual control mode – whether a trait is influenced by the offspring or maternal genotype alone or by both jointly. In Refs [64,65], a similar approach was proposed in which the interactions between pairs of loci in separate genomes can be fully studied, like a two-dimensional search. In doing this, the endosperm genotypes were not nested but completely crossed with the maternal ones to produce more combinations of joint genotypes. The computation was made feasible by adopting a more precise two-stage hierarchical design that integrates marker genotypes from both the maternal plants and embryos. The method was later expanded by the authors to include maternal–embryo and embryo–endosperm epistasis, leading to a novel strategy called genome–genome epistasis mapping (Figure 1) [66]. Such a triple genome interaction model provides valuable insights into the genetic mechanisms involved in the complex three-way crosstalk that coordinates the development of seeds. A variation under the Mendelian radar: gene imprinting However the endosperm arose in flowering plants, the uneven contribution of genomes from the male and female parents is associated with differences in expression of their genes, and this can have profound effects on endosperm and embryo development. Endosperms that vary from a 2C:1C maternal to paternal genome ratio generally fail, resulting in embryo abortion. The molecular basis for this trait, the socalled endosperm balance number, or EBN [67], is not understood. It could be related to the ‘parental conflict theory’ [68], which explains that there is a difference in the reproductive strategy of the male and female parents. Flowering plants, like mammals, nurture their offspring 395

Review

Figure 1. (a) The landscapes of the log-likelihood ratio (LR) test statistics calculated for the hypothesis test for the existence of a QTL against the maternal and endosperm QTL locations on the assumed linkage group when the two QTLs are located at the same marker interval (heritability of 0.4). Adapted from [65]. (b) Profiles of the log-likelihood ratio (LR) test statistics for testing a genome–genome epistatic QTL affecting mean ploidy based on a joint analysis of the four backcross populations, calculated as a function of the endosperm and embryo genome positions across the linkage map. The arrows indicate the two peaks of the LR profiles at the locations of the endosperm (on chromosomes 4 and 10) and embryo QTLs (on chromosome 9) that display a significant epistatic effect. Adapted from [78].

through accessory tissues (endosperm or placenta, respectively). Although the offspring arise from a single female parent, there are often different male parents. The male benefits if a larger proportion of nutrients is directed to his offspring, whereas the female benefits if nutrients are equally distributed to all her progeny. Consequently, genes expressed from the male genome enhance nutrient uptake, whereas genes expressed from the female genome limit nutrient uptake and the rate of endosperm development. The physiological processes underlying these responses are unknown, but parent-of-origin-specific allele expression has been demonstrated in endosperm through the process of ‘gene imprinting’. The imprinting process involves DNA methylation, which affects chromatin structure and, ultimately, gene transcription. Gene imprinting is currently a subject of intense interest and research [5,69–71]. Gene imprinting has been thought to have an important role in modulating endosperm cell number, endoreduplication (and thus grain size), grain filling and other seed 396

Trends in Plant Science Vol.14 No.7

quality traits [72]. Although several imprinted genes, such as MEA (MEDEA) and FIS2 (FERTILIZATION INDEPENDENT SEED2), have been identified in Arabidopsis endosperm, little is known about the true scope of imprinting and its contribution to the quantitative variation and genetic architecture of complex seed traits. The imprinting effect has been a blind spot of standard quantitative genetic theory owing to its violation of the usual ‘euMendelian symmetries’ [73]. Asymmetric expression means that the genotypic values should be assigned based on the origins of alleles. Genetic mapping of both Mendelian QTLs and imprinted QTLs (iQTLs) have received considerable attention. But most approaches were restricted to controlled outbred crosses, which are often subject to confounding effects caused by high heterozygosity at a given locus. The common F2 mapping design, however, does not have adequate degrees of freedom to distinguish the two parent-of-origin-dependent forms of the heterozygote (i.e. Qq and qQ). A workable model was proposed in F2 design [74] by capitalizing on the sex-specific difference in recombination between male and female chromosomes, in which the same alleles from different parents can be traced. Another effective design, reciprocal backcross, has been suggested for iQTL mapping, which eliminates the need to rely totally on the sex-specific recombination fraction [75,76]. Two forms of parameterization to measure the probability of imprinting have also been introduced [75]. The estimation and hypothesis testing of these parameters can determine not only the existence of imprinting but also its degree (complete or partial) and type (maternal or paternal). A realistic endosperm QTL model should not ignore maternal effects, which can be a confounding factor of imprinting. Recently, a unified approach for estimating maternal–zygotic interactions and imprinting effects of QTLs has been proposed [77] and has been successfully applied to study endoreduplication in maize endosperm. Nevertheless, the interplay among imprinting, maternal effects and other confounding factors, such as genotype-byenvironment interaction, is much more complicated in seeds and requires further research. Prospects The endosperm of seeds can serve as a valuable system for addressing fundamental questions related to the improvement of seed size in crops and the developmental biology of plants. Although many QTLs that control endosperm traits have been identified, the mechanisms that control endosperm formation and development are poorly understood. Coordinated exchanges of signals among the endosperm, the embryo and the maternal integument are crucial for proper seed development. Genetic mapping based on a reciprocal mating design allows the characterization of how maternal, embryo and endosperm QTLs interact to regulate endosperm development [77]. This design, implemented with statistical and computational algorithms, enables researchers to quantify and test genetic imprinting effects on QTLs and the coordination of QTLs in different parts of a seed. Genetic studies of endosperm can be improved through the use of functional genetic mapping, an approach that detects the temporal expression pattern of dynamic QTLs

Review during endosperm development. Functional mapping is able to determine when a particular endosperm QTL turns on and off and its pleiotropic effects. This information is valuable for understanding the interplay between gene actions and the program and pattern of endosperm development. Since the introduction of gene microarrays, there has been considerable interest in using whole-genome expression profiling to gain insight into plant traits and to identify their genetic mediators. An emerging systems biology approach has to the potential to reveal the complex relationships of gene interactions and to identify the pathways that mediate, directly or indirectly, the dynamic behavior of gene networks. Such an approach could be used to elucidate the genetic architecture of endosperm traits. Furthermore, techniques have been developed to detect regulatory control of each step from DNA to life. Metabolites are the end product of nearly all cellular processes and reflect the outcome of changes directed by genetic and proteomic adjustments to environmental stimuli and genetic modifications. A complete explanation of the nature of seed growth and development will require integration of genomic, transcriptomic, proteomic and metabolomic data. This systems biology approach will provide the opportunity to study the dynamics of gene networking and describe the complete physiology of endosperm development. Acknowledgements C.X. is funded by the National Basic Research Program of China (grant no. 2006CB101700), the National Natural Science Foundation of China (grant no. 30370758) and the Program for New Century Excellent Talents in University of the Ministry of Education of China. R.W. is funded by a joint National Science Foundation/National Institutes of Health grant (DMS/NIGMS-0540745). X.W. is supported by a Merck Foundation Quantitative Science Fellowship. B.A.L. gratefully acknowledges research support from the US Department of Energy (grant no. DEFG05–95ER20194). We thank three anonymous reviewers for helpful suggestions.

References 1 Lopes, M.A. and Larkins, B.A. (1993) Endosperm origin, development, and function. Plant Cell 5, 1383–1399 2 Becraft, P.W. (2001) Cell fate specification in the cereal endosperm. Semin. Cell Dev. Biol. 12, 387–394 3 Olsen, O-A. (ed.) (2007) Endosperm: Developmental and Molecular Biology, Springer-Verlag 4 Sabelli, P.A. and Larkins, B.A. (2009) The development of endosperm in grasses. Plant Physiol. 149, 14–26 5 Friedman, W.E. et al. (2008) Origin of the fittest and survival of the fittest: relating female gametophyte development to endosperm genetics. Int. J. Plant Sci. 169, 79–92 6 Capitanio, R. et al. (1983) Grain weight and its components in maize inbred lines. Maydica XXVIII, 365–379 7 Reddy, V.M. and Daynard, T.B. (1985) Endosperm characteristics associated with rate of grain filling and kernel size in corn. Maydica XXVIII, 339–355 8 Olsen, O-A. (2001) Endosperm development: cellularization and cell fate specification. Annu. Rev. Plant Physiol. 52, 233–267 9 Larkins, B.A. et al. (2001) Investigating the hows and whys of DNA endoreduplication. J. Exp. Bot. 52, 183–192 10 Friedman, W.E. (2001) Developmental and evolutionary hypotheses for the origin of double fertilization and endosperm. C.R. Acad. Sci. III 324, 559–567 11 Sargant, E. (1900) Recent work on the results of fertilization in angiosperms. Ann. Bot. (Lond.) 14, 689–712 12 Friedman, W.E. (1990) Double fertilization in Ephedra, a nonflowering seed plant: its bearing on the origin of angiosperms. Science 247, 951–954

Trends in Plant Science

Vol.14 No.7

13 Carmichael, J.S. and Friedman, W.E. (1995) Double fertilization in Gnetum gnemon: the relationship between the cell cycle and sexual reproduction. Plant Cell 7, 1975–1988 14 Nowack, M.K. et al. (2007) Bypassing genomic imprinting allows seed development. Nature 447, 312–315 15 Ungru, A. et al. (2008) Natural variation in the degree of autonomous endosperm formation reveals independence and constraints of embryo growth during seed development in Arabidopsis thaliana. Genetics 179, 829–841 16 Lander, E.S. and Botstein, D. (1989) Mapping mendelian factors underlying quantitative traits using RFLP linkage maps. Genetics 121, 185–199 17 Wu, R. et al. (2002) An improved genetic model generates highresolution mapping of QTL for protein quality in maize endosperm. Proc. Natl. Acad. Sci. U. S. A. 99, 11281–11286 18 Xu, C. et al. (2003) Mapping quantitative trait loci underlying triploid endosperm traits. Heredity 90, 228–235 19 Wu, R. et al. (2002) Statistical methods for dissecting triploid endosperm traits using molecular markers: an autogamous model. Genetics 162, 875–892 20 Kao, C-H. (2004) Multiple-interval mapping for quantitative trait loci controlling endosperm traits. Genetics 167, 1987–2002 21 Dempster, A. et al. (1977) Maximum likelihood from incomplete data via the EM algorithm. J. R. Stat. Soc. [Ser A] 39, 1–38 22 Tan, Y.F. et al. (1999) The three important traits for cooking and eating quality of rice grains are controlled by a single locus in an elite rice hybrid, Shanyou 63. Theor. Appl. Genet. 99, 642–648 23 Wang, X. and Larkins, B.A. (2001) Genetic analysis of amino acid accumulation in opaque-2 maize endosperm. Plant Physiol. 125, 1766– 1777 24 Wang, X. et al. (2001) Quantitative trait locus mapping of loci influencing elongation factor 1alpha content in maize endosperm. Plant Physiol. 125, 1271–1282 25 Wang, X. et al. (2008) A mixture model approach to the mapping of QTL controlling endosperm traits with bulked samples. Genetica 132, 59–70 26 Wen, Y. and Wu, W. (2006) Methods for mapping QTLs underlying endosperm traits based on random hybridization design. Chin. Sci. Bull. 51, 1976–1981 27 Wen, Y. and Wu, W. (2008) Experimental designs and statistical methods for mapping quantitative trait loci underlying triploid endosperm traits without maternal genetic variation. J. Hered. 99, 546–551 28 Wang, X. et al. (2009) Improved genetic mapping of endosperm traits using NCIII and TTC designs. J. Hered, DOI: 10.1093/jhered/esp009 29 Mauricio, R. (2001) Mapping quantitative trait loci in plants: uses and caveats for evolutionary biology. Nat. Rev. Genet. 2, 370–381 30 Zeng, Z.B. (1993) Theoretical basis for separation of multiple linked gene effects in mapping quantitative trait loci. Proc. Natl. Acad. Sci. U. S. A. 90, 10972–10976 31 Li, H. et al. (2007) A modified algorithm for the improvement of composite interval mapping. Genetics 175, 361–374 32 Kao, C-H. et al. (1999) Multiple interval mapping for quantitative trait loci. Genetics 152, 1203–1216 33 Tibshirani, R. (1996) Regression shrinkage and selection via the lasso. J. Roy. Statist. Soc. Ser. B. Methodological 58, 267–288 34 Yi, N. and Xu, S. (2008) Bayesian LASSO for quantitative trait loci mapping. Genetics 179, 1045–1055 35 Tibshirani, R. et al. (2005) Sparsity and smoothness via the fused lasso. J. R. Stat. Soc. B 67, 91–108 36 Yuan, M. and Lin, Y. (2006) Model selection and estimation in regression with grouped variables. J. R. Stat. Soc. B 68, 49–67 37 Jiang, C. and Zeng, Z. (1995) Multiple trait analysis of genetic mapping for quantitative trait loci. Genetics 140, 1111–1127 38 Knott, S.A. and Haley, C.S. (2000) Multitrait least squares for quantitative trait loci detection. Genetics 156, 899–911 39 Xu, C. et al. (2005) Joint mapping of quantitative trait loci for multiple binary characters. Genetics 169, 1045–1059 40 Li, R. et al. (2006) Structural model analysis of multiple quantitative traits. PLoS Genet. 2, e114 41 Banerjee, S. et al. (2008) Bayesian quantitative trait loci mapping for multiple traits. Genetics 179, 2275–2289 42 Becraft, P.W. et al. (1996) CRINKLY4: a TNFR-like receptor kinase involved in maize epidermal differentiation. Science 273, 1406–1409

397

Review 43 Gifford, M.L. et al. (2003) The Arabidopsis ACR4 gene plays a role in cell layer organisation during ovule integument and sepal margin development. Development 130, 4249–4258 44 Ma, C-X. et al. (2002) Functional mapping of quantitative trait loci underlying the character process: a theoretical framework. Genetics 161, 1751–1762 45 Wu, R. et al. (2004) A general framework for analyzing the genetic architecture of developmental characteristics. Genetics 166, 1541–1551 46 Wu, R. and Lin, M. (2006) Functional mapping-how to map and study the genetic architecture of complex dynamic traits. Nat. Rev. Genet. 7, 229–237 47 Schweizer, L. et al. (1995) Dynamics of maize endosperm development and DNA endoreduplication. Proc. Natl. Acad. Sci. U. S. A. 92, 7070–7074 48 Garcia, D. et al. (2005) Maternal control of integument cell elongation and zygotic control of endosperm growth are coordinated to determine seed size in Arabidopsis. Plant Cell 17, 52–60 49 Grini, P.E. et al. (2002) Embryo and endosperm development is disrupted in the female gametophytic capulet mutants of Arabidopsis. Genetics 162, 1911–1925 50 Chaudhury, A.M. and Berger, F. (2001) Maternal control of seed development. Semin. Cell Dev. Biol. 12, 381–386 51 Pagnussat, G.C. et al. (2005) Genetic and molecular identification of genes required for female gametophyte development and function in Arabidopsis. Development 132, 603–614 52 Berger, F. et al. (2006) Endosperm: an integrator of seed growth and development. Curr. Opin. Plant Biol. 9, 664–670 53 Birchler, J.A. (1993) Dosage analysis of maize endosperm development. Annu. Rev. Genet. 27, 181–204 54 Dilkes, B.P. et al. (2002) Genetic analyses of endoreduplication in Zea mays endosperm evidence of sporophytic and zygotic maternal control. Genetics 160, 1163–1177 55 Colombo, L. et al. (1997) Downregulation of ovule-specific MADS box genes from petunia results in maternally controlled defects in seed development. Plant Cell 9, 703–715 56 Shi, C. et al. (1997) Analysis of genetic effects for nutrient quality traits in indica rice. Theor. Appl. Genet. 92, 1099–1102 57 Zheng, X. et al. (2008) The QTL analysis on maternal and endosperm genome and their environmental interactions for characters of cooking quality in rice (Oryza sativa L.). Theor. Appl. Genet. 116, 335–342 58 Berger, F. (2003) Endosperm: the crossroad of seed development. Curr. Opin. Plant Biol. 6, 42–50 59 Cockerham, C.C. and Weir, B.S. (1977) Quadratic analyses of reciprocal crosses. Biometrics 33, 187–203 60 Foolad, M. and Jones, R. (1992) Models to estimate maternally controlled genetic variation in quantitative seed characters. Theor. Appl. Genet. 83, 360–366 61 Zhu, J. and Weir, B.S. (1994) Analysis of cytoplasmic and maternal effects. II. Genetic models for triploid endosperms. Theor. Appl. Genet. 89, 160–166

398

Trends in Plant Science Vol.14 No.7 62 Hu, Z. and Xu, C. (2005) A new statistical method for mapping QTLs underlying endosperm traits. Chin. Sci. Bull. 50, 1470–1476 63 Hu, Z. et al. (2006) A method for identification of the expression mode and mapping of QTL underlying embryo-specific characters. J. Hered. 97, 473–482 64 Cui, Y. et al. (2004) Mapping quantitative trait loci interactions from the maternal and offspring genomes. Genetics 167, 1017–1026 65 Cui, Y. and Wu, R. (2005) Statistical model for characterizing epistatic control of triploid endosperm triggered by maternal and offspring QTLs. Genet. Res. 86, 65–75 66 Cui, Y. and Wu, R. (2005) Mapping genome–genome epistasis: a highdimensional model. Bioinformatics 21, 2447–2455 67 Johnston, S. et al. (1980) The significance of genetic balance to endosperm development in interspecific crosses. Theor. Appl. Genet. 57, 5–9 68 Haig, D. and Westoby, M. (1989) Parent-specific gene expression and the triploid endosperm. Am. Nat. 134, 147–155 69 Vinkenoog, R. et al. (2003) Genomic imprinting and endosperm development in flowering plants. Mol. Biotechnol. 25, 149–184 70 Kinoshita, T. et al. (2008) Genomic imprinting: a balance between antagonistic roles of parental chromosomes. Semin. Cell Dev. Biol. 19, 574–579 71 Garnier, O. et al. (2008) Genomic imprinting in plants. Epigenetics 3, 14–20 72 Khachatourians, G.G. et al. (2002) Transgenic Plants and Crops. CRC Press 73 Spencer, H. (2009) Effects of genomic imprinting on quantitative traits. Genetica 136, 285–293 74 Cui, Y. et al. (2006) Model for mapping imprinted quantitative trait loci in an inbred F2 design. Genomics 87, 543–551 75 Cui, Y. et al. (2007) A statistical model for dissecting genomic imprinting through genetic mapping. Genetica 130, 227–239 76 Cui, Y. (2007) A statistical framework for genome-wide scanning and testing of imprinted quantitative trait loci. J. Theor. Biol. 244, 115–126 77 Li, Y. et al. (2008) A statistical model for estimating maternal–zygotic interactions and parent-of-origin effects of QTLs for seed development. PLoS One 3, e3131 78 Coelho, C.M. et al. (2007) Identification of quantitative trait loci that affect endoreduplication in maize endosperm. Theor. Appl. Genet. 115, 1147–1162 79 Carlson, E. (2004) Mendel’s Legacy: The Origin of Classical Genetics. CSHL Press 80 Mo, H. (1987) Genetic expression for endosperm traits. In Proceedings of the Second International Conference on Quantitative Genetics: 1987 May 30–June 5; Raleigh, NC, USA (Weir, B. et al., eds), pp. 478–487, Sinauer Associates 81 Bogyo, T. et al. (1988) Genetic models for quantitatively inherited endosperm characters. Heredity 60, 61–67 82 Wen, Y. and Wu, W. (2007) Interval mapping of quantitative trait loci underlying triploid endosperm traits using F3 seeds. J. Genet. Genomics 34, 429–436