- American Society of Plant Biologists
Abstract
This review describes recent advances in the analysis of metabolism using quantitative genetics. It focuses on how recent metabolic quantitative trait loci (QTL) studies enhance our understanding of the genetic architecture underlying naturally variable phenotypes and the impact of this fundamental research on agriculture, specifically crop breeding. In particular, the role of whole-genome duplications in generating quantitative genetic variation within a species is highlighted and the potential uses of this phenomenon presented. Additionally, the review describes how new observations from metabolic QTL mapping analyses are helping to shape and expand the concepts of genetic epistasis.
INTRODUCTION
Biochemistry and genetics have long been entwined in a mutualistic relationship beginning with Mendel's reliance upon metabolic phenotypes (anthocyanins and starch) to develop basic genetic theory. Biochemistry has advanced the understanding of DNA structure and metabolism, and genetic analyses have been equally crucial for the construction of biochemical pathways (Tatum et al., 1950; Watson and Crick, 1953; Yanofsky, 1960). The relationship between biochemistry and genetics has yielded great benefits to the broader scientific community. For instance, the analysis of genetic mutants in the Escherichia coli lactose and Trp biosynthetic pathways have shaped our current understanding of gene regulation (Yanofsky, 1960; Jacob and Monod, 1961). Even today, the integration of genetic and modern metabolic network theory are forging connections between genome-scale data and organismal phenotypes (Segre et al., 2005; Gjuvsland et al., 2007; Fu et al., 2009).
The combination of biochemistry and genetics has been of great utility in the study of plants. Biochemical phenotypes generated by defects in processes such as flavonol and Trp biosynthesis have been fundamental to developing basic genetic approaches within model plants (Koornneef et al., 1982; Last and Fink, 1988; Shirley et al., 1995). Genetic studies have also identified biochemical processes that proved recalcitrant to solely biochemical approaches (Humphreys and Chapple, 2002; Xie et al., 2003). Ecological and evolutionary genetic studies of plants have relied strongly on biochemical phenotypes, predominantly those manifested as variation in flower color, to further our understanding of how both biochemistry and genetics control fitness in wild plant populations (Brown and Clegg, 1984; Stanton et al., 1991; Holton et al., 1993).
The combination of biochemistry and genetics can also provide powerful insights into the origins and maintenance of natural variation. Modern biochemical analyses allow quantification of metabolic diversity that can be associated with specific genetic markers, mRNA transcripts, and enzyme activities, allowing linkage between variation from genetic to biochemical levels that is more complicated for less-defined or more pleiotropic phenotypes, such as growth or disease susceptibility. The ability to analyze variation at different functional levels provides an opportunity to query the mechanisms by which this variation evolves and is maintained in populations and species.
Recent advances in genome sequencing and high-throughput phenotyping technologies allow plant biologists to revisit the study of quantitative genetics and how genetic variation is connected to phenotypic variation. The combination of biochemistry with quantitative genetics has a long history in plants, as one of the first identified QTL controls a biochemical phenotype, specifically, seed color in Phaseolus (Sax, 1923). More recently, these combined disciplines have expanded our fundamental understanding of how metabolites can provide biotic stress resistance in both Zea mays and Brassica (Mithen and Magrath, 1992; Magrath et al., 1993; Mithen and Toroser, 1995; Byrne et al., 1996, 1998; McMullen et al., 1998). The advent of broad-spectrum metabolite profiling technologies is enabling a fuller investigation into the influence of quantitative genetic variation on the plant metabolome and correspondingly how the fitness consequences of these metabolic changes alter the genetic architecture of the species. Several previous reviews cover the use of metabolic QTL to integrate across different levels of genomic information (sequence, transcript, and protein) to better understand biomass production, improve crop breeding, and obtain ecological inference about the corresponding selective pressure acting on these QTL (Fernie and Schauer, 2009; Keurentjes, 2009; Kliebenstein, 2009). These reviews also provide significant evidence for the myriad specific molecular events that can lead to altered phenotypes, including epigenetic, coding sequence, and regulatory polymorphisms. This review focuses on how recent metabolic QTL studies enhance our fundamental understanding of the genetic architecture underlying naturally variable phenotypes and the impact of this basic research on agriculture, specifically crop breeding.
METABOLIC QTL AND THE UNDERPINNINGS OF NATURAL GENETIC VARIATION
At a basic level, we presume that we understand the very underpinnings of qualitative and consequentially Mendelian genetics, concepts such as allelic dominance, heritability, and epistasis (Table 1 ). However, all of these concepts exist on quantitative scales, and there is little knowledge about the distributions of these genetic forces. Understanding these distributions and any relationship between them is essential to developing predictive approaches for the analysis of natural genetic variation, irrespective of the field of study. Due to its relative economy, metabolomics or metabolic profiling is becoming a popular phenotyping platform for quantitative genetic analysis of biochemistry at a genome-wide level.
How Genetics Terms Are Defined within This Article
In a multiyear study of tomato (Solanum lycopersicum) metabolism, 76% of metabolic QTL displayed a dominance relationship between the alleles, whereas 18% of the metabolic QTL had a semidominant relationship (Schauer et al., 2006, 2008). Interestingly, the distribution of dominance effects varied between metabolic classes, with sugar alcohol-linked QTL showing the lowest proportion of semidominance and organic acids showing the highest proportion of semidominance. This suggests that different phenotypic classes can have dissimilar dominance relationships among naturally variable alleles, possibly due to their different underlying molecular networks. Within gene expression QTL, cis-acting expression polymorphisms show greater additive effects, whereas trans-acting polymorphisms are more likely to show a dominant effect. One explanation for this discrepancy is that a cis-acting polymorphism by definition cannot affect the other allele in a heterozygous individual and thereby prevent it from being dominant. It remains to be seen if this relationship differs between networks or related phenotypes as with metabolic traits (Stupar and Springer, 2006; Lemos et al., 2008).
Heritability may also differ among classes of metabolic traits. The majority of metabolite studies estimate broad-sense heritability (H2), which combines all potential genetic and epigenetic contributions to the phenotype. Using this approach, an analysis of tomato metabolites showed that H2 significantly varied among metabolites, revealing distinct clusters of high and low heritability metabolites (Schauer et al., 2008). This observation was echoed by an analysis of Arabidopsis thaliana secondary metabolites showing that two different classes of secondary metabolite, indolic and aliphatic glucosinolates, detected on the same phenotyping platform had distinctly different H2 estimates in multiple mapping populations (Kliebenstein et al., 2001b; Wentzell et al., 2007). Interestingly, the transcripts encoding the biosynthetic enzymes for these metabolites had higher estimated heritability than the metabolites themselves (Wentzell et al., 2007; West et al., 2007). Thus, it is clear that heritability (e.g., genetic variation) can differ across molecular levels: transcript, metabolite, and physiology, for what is typically considered as the same phenotype, processes involved in glucosinolate accumulation. This difference in heritability could be due to each higher-order phenotype (transcript, metabolite, and physiology) integrating more and different inputs than the lower-order phenotypes. For instance, larger or more complex regulatory networks may affect a phenotype at the metabolic level than at the transcript level. Alternatively, each additional phenotypic level may accumulate additional stochastic noise, thus decreasing the estimate of heritability (Elowitz et al., 2002; Raser and O'Shea, 2004). Broader tests of measurable phenotypes in multiple species and mapping population are required to separate these possibilities.
Application
A major goal of combining genomic analysis with quantitative genetics applications, such as plant breeding, is to allow a predictive connection from genetic to phenotypic variation. This would allow crop developers to use genomic information efficiently to engineer a phenotypic change within the crop. However, this requires a deep understanding of basic genetic components like heritability and allelic dominance. Understanding how these basic variables differ among phenotypes will allow genomic information to be incorporated more efficiently into breeding schemes. A complication of this application is that broad-sense heritability is less useful for breeding applications than narrow-sense heritability, which is the genetic variation that additively contributes to the phenotype (Table 1). Broad- and narrow-sense heritability estimates can differ dramatically depending on the level of allelic dominance, multilocus epistasis, and parental influence on the phenotype (Falconer and Mackay, 1996; Lynch and Walsh, 1998). Studies have not yet thoroughly queried the relationship between these two measures of heritability on a genomic scale, which will be required to enable the full use of genomics to predict phenotypic consequences and allow for predictive engineering of crops.
METABOLIC QTL AND WHOLE-GENOME DUPLICATIONS
The study of metabolic QTL may shed light on evolutionary questions regarding the fate of duplicated genes. If a duplication event produces two copies of a gene that are afterwards maintained, these duplicated genes may diverge from the ancestral gene or from each other. This divergence can occur via neofunctionalization, where one of the duplicates obtains a novel function, or subfunctionalization, such that the duplicate copies obtain differential expression patterns in terms of tissue specificity or stress response (Ohno, 1970; Lynch and Conery, 2000). Interestingly, most duplicates from whole-genome duplications are lost over long time frames as represented by speciation events (Freeling, 2009), yet within species, these duplicates appear to play significant roles in QTL formation and maintenance (Kliebenstein, 2008). While further work is necessary to test if this is a biological discrepancy or an observational bias, the analysis of metabolic QTL has both identified examples of subfunctionalization and neofunctionalization and identified new roles for these evolutionary mechanisms within a species.
Neofunctionalization and Metabolic Diversity
The cloning of metabolic QTL, largely for secondary metabolites, has revealed several examples of neofunctionalization where duplicated genes have evolved independent biochemical functions (Kliebenstein et al., 2001a; Lambrix et al., 2001; de Meijer et al., 2003; Benderoth et al., 2006; Schnee et al., 2006; Burow et al., 2009). For example, a tandem duplication generated two genes in glucosinolate metabolism, AOP2 and AOP3, that evolved different biochemical reactions using the same substrate (Kliebenstein et al., 2001a). AOP2 maintained the ancestral function of converting a methylsulfinyl moiety to an alkenyl, while AOP3 derived the ability to convert the methylsulfinyl to a hydroxyalkyl. Showing that QTL controlling metabolic diversity in secondary metabolites result directly from neofunctionalization suggests that gene duplication may facilitate plant evolution of new responses to a complex and changing environment, partly through increasing metabolic diversity (Ober, 2005). This agrees with recent work showing that duplicated genes are overrepresented in responses to biotic and abiotic stress (Hanada et al., 2008; Kliebenstein, 2008).
Subfunctionalization and Expression Diversity
Duplicate genes may also subfunctionalize, developing expression patterns that differ spatially or temporally from ancestral expression patterns. A hallmark of subfunctionalization in metabolic QTL analysis is detection of tissue-specific QTL for an enzymatic activity, which is likely to have resulted from differential expression patterns between members of an encoding gene family (Sergeeva et al., 2004). These analyses can be extended to identify the genes controlling the biochemical reaction in specific tissues (Sergeeva et al., 2006). However, more detailed metabolic QTL experiments are beginning to show that gene duplicates diverging by tissue expression pattern may represent only the simplest scenario, while subfunctionalization can generate complex expression patterns that integrate development, ontogeny, and environment such that duplicated genes may show highly conditional effects (Wentzell et al., 2008; Wentzell and Kliebenstein, 2008; Burow et al., 2009). In one example, the main metabolic QTL controlling secondary metabolite-mediated insect resistance within Arabidopsis, ESP and ESM1, result from two successive whole-genome duplications generating four highly similar genomic regions (Wentzell et al., 2008). Further endoreduplication and gene loss accompanied by promoter divergence has generated QTL conditional on tissue, ontogeny, and environmental influences (Lambrix et al., 2001; Zhang et al., 2006; Wentzell and Kliebenstein, 2008; Burow et al., 2009). Gene duplication, followed by sequence changes that altered expression patterns of duplicated genes, has apparently given the plant more precise control over its metabolic defenses. As such, subfunctionalization may be more prevalent in biotic stress responses. Given that the above observations pertain to precise regulatory variation among diverse genotypes, this highly integrated regulation would be difficult to study using single gene mutants from a single reference genotype but is readily accessible in natural populations. The frequency of these higher-order interactions remains to be determined.
Genetic Subfunctionalization
In addition to obtaining new functions or novel partitioning of function, metabolic QTL studies suggest that duplicate genes can obtain independent loss-of- or decrease-in-function polymorphisms, essentially a genetic subfunctionalization where genes have different levels of function in different genotypes. The simplest form of this genetic subfunctionalization is exemplified by the recent cloning of two enzymatic QTL that combine to form a lethal epistatic interaction (Bikard et al., 2009). In this example, a gene encoding a key enzyme for His biosynthesis, HPA, was duplicated within Arabidopsis and then divergent accessions independently lost one or the other copy of the gene. F2 progeny generated from crosses of these divergent accessions showed segregation patterns where one-sixteenth of the progeny lacked both copies of the HPA gene and were thus nonviable (Bikard et al., 2009). A secondary metabolism family of flavin-monooxygenases shows hallmarks of genetic subfunctionalization (Li et al., 2008). However, it is still unknown how frequently genetic subfunctionalization may occur within larger gene families, especially given that gene duplicates show elevated levels of transcript abundance polymorphism (Kliebenstein, 2008).
Application
Understanding the complex contributions of gene duplication to plant metabolism may benefit efforts to clone individual metabolic QTL and provide precise markers for marker-assisted selection. Essentially, identification and cloning of a QTL for a given trait in a single tissue, condition, or genotype implicates homoeologous members of the same gene family that colocalize with the other QTL as candidate genes for control of the trait in other tissues or conditions (Figure 1 ). For an example, take a species where a gene has been triplicated via a whole-genome duplication and an ensuing segmental duplication of only one of the resulting copies (Figure 1). If a researcher identified the molecular basis of QTLa that controls a phenotype in the shoot and root, the homoeologous duplicates of this gene might be considered strong candidates for control of root-specific QTL. As such, cloning one QTL would potentially accelerate identification of the further subfunctionalized duplicate genes with minimal investment of new resources. More broadly, whole-genome assessments of duplication patterns (Vision et al., 2000) could be analyzed in concert with genomic positions of QTL for a given trait to identify relationships between QTL incidence and genome duplication events.
Homoeologous QTL.
Two hypothetical QTL results are shown for the same phenotype in the leaf and root. Peak locations represent the approximate genetic positions of the loci, with small letters labeling the three loci. The bar at the bottom shows the position of the three corresponding genomic regions arising via a triplication of a single original genomic fragment.
METABOLIC QTL AND MOLECULAR MECHANISMS UNDERLYING GENETIC EPISTASIS
Epistasis is loosely defined as an interaction between alleles at two or more genes, and there is long-standing debate about the presence and relevance of epistasis in genetic studies (Fisher, 1930; Wright, 1931; Carlborg and Haley, 2004; Hill et al., 2008). In functional molecular studies, epistasis generally implies a regulatory or physical interaction between the genes and their products, specified here as functional epistasis (Phillips, 2008). By contrast, epistasis in genetic studies indicates that the link between allelic variation and phenotypic variation at one locus is dependent upon the allelic state at another locus or loci, specified here as genetic epistasis. In a quantitative genetics study, epistasis appears as a statistical interaction between loci that does not allow for immediate divination of the molecular causes.
One area where epistatic interactions between naturally variable loci is of key importance is the study of genome incompatability in Arabidopsis, Mimulus spp, and other plants (Fishman and Willis, 2001; Bomblies et al., 2007; Sweigart et al., 2007; McDaniel et al., 2008). These are often loosely classified as Dobzhansky-Muller incompatibility loci whereby the two genes have diverged in function such that when reintroduced they cause a hybrid incompatibility that may be associated with speciation events (Dobzhansky, 1937; Muller, 1942). Furthermore, there is significant interest in the analysis of hybrid necroses, incompatibility combinations that alter plant defense mechanisms (Bomblies et al., 2007; Alcazar et al., 2009). Genomic incompatibilities are not phenotypically limited but can result from any multilocus combination of alleles that generates negative effects on diverse phenotypes, such as chloroplast development, disease resistance, and pollen viability (Fishman and Willis, 2001; Bomblies et al., 2007; Sweigart et al., 2007; McDaniel et al., 2008). Similar arguments have been made to explain the positive impacts of genome mixing upon heterosis or hybrid vigor (Birchler et al., 2003). However, current data are insufficient to connect observations of genetic epistasis with molecular mechanisms consistent with solely a functional epistasis explanation.
It is in the arena of explaining the basis of genetic epistasis that the analysis of metabolic QTL displays its power to address basic genetic questions while remaining useful in applied settings such as crop breeding. Metabolic QTL analyses are beginning to identify three broad classes of molecular mechanism underlying genetic epistases in natural populations: functional epistasis, gene family epistasis, and diffuse epistasis. The following sections define these three postulated forms of epistasis and provide examples identified via metabolic QTL studies.
Functional Epistasis and Metabolite-Transcription Factor Communication
The most familiar form of epistasis is functional epistasis, where genes are connected within a biochemical or regulatory pathway. Cloned metabolic QTL reveal numerous instances where genetic epistasis is controlled by traditional functional epistasis. For example, in the maize (Zea mays) flavone biosynthetic pathway, multiple regulatory loci interact to determine the level of this defensive compound (Byrne et al., 1996, 1998; McMullen et al., 1998). In agreement, regulatory interactions that simultaneously and directly control transcriptional, enzymatic, and metabolite levels for a specific metabolite or pathway have been identified within Arabidopsis (Fu et al., 2009). These studies demonstrate how standard regulatory or enzymatic interactions can generate genetic epistasis through functional epistasis. However, there are also numerous instances where a given QTL does not control all three levels, suggesting that there are complex and potentially different networks controlling these three levels (Keurentjes et al., 2006, 2007, 2008).
One such potential complexity is that epistatic interactions are not limited to functional networks involving simple linear pathways. The Arabidopsis glucosinolate biosynthetic pathway presents an advanced example of the link between genetic and functional epistasis. Six QTL controlling glucosinolate accumulation show genetic epistasis (Kliebenstein et al., 2005; Kliebenstein, 2009) (Figure 2A ). These six loci act together to control both the type and relative level of glucosinolate accumulation, and the underlying genes encode four enzymes and two transcription factors (Kliebenstein et al., 2005; Sønderby et al., 2007; Hansen et al., 2008; Kliebenstein, 2009) (Figure 2B). This work showed that the two transcription factor loci, MYB28 and MYB29, directly bind promoters to regulate transcript abundance of the four enzyme-encoding genes, GS-OX, AOP2, GS-OH, and MAM1, as well as other genes encoding glucosinolate biosynthetic enzymes (Gigolashvili et al., 2007, 2008; Hirai et al., 2007) (Figure 2B). Whereas most models of functional epistasis are sufficiently explained with transcription factors controlling enzymes, the glucosinolate QTL analyses showed that enzyme-encoding QTL altered transcript abundance for both the two MYB transcription factors and the other biosynthetic genes, suggesting an effect of these enzymes upon transcript levels (Kliebenstein et al., 2006; Wentzell et al., 2007). Transgenic introduction of one of these enzymes, AOP2, into a null background altered transcript levels for the transcription factors as well as the biosynthetic pathway, suggesting that one of the metabolic products of AOP2 can modulate transcript levels of the MYB28 and MYB29 transcription factors (Wentzell et al., 2007). Thus, this complex genetic epistasis involves signaling loops whereby transcription factors control expression of the biosynthetic pathway that in turn affect transcript accumulation of the transcription factors (Sønderby et al., 2007; Wentzell et al., 2007) (Figure 2). As such, the analysis of metabolic QTL is providing a richer image of functional epistasis than traditional linear relationship assumptions have allowed.
Functional Epistasis Underlying Complex Genetic Epistasis.
(A) The known genetic epistasis linking six cloned QTL controlling glucosinolate accumulation within Arabidopsis is shown. The names of the QTL are shown with lines indicating genetic epistatic interactions identified within any of the existing natural Arabidopsis populations.
(B) The current model of functional epistasis underlying the above genetic epistasis is shown. Small arrows show specific enzymes with names showing those enzymes that are the mechanistic basis of specific QTL in (A). Metabolic structures show the specific metabolites that accumulate from these enzymes. The line arrows show the proposed link between a glucosinolate metabolite and the control of transcription for the MYB28/MYB29 transcription factors. The double question mark shows that this is just one possible connection among many metabolite/gene interactions. The circles represent the MYB28/MYB29 proteins with line arrows showing the promoters with which these transcription factors interact. These genes encode the enzymes shown by the arrows for the metabolic pathway.
Genetic Epistasis within Gene Families
In contrast with functional epistasis, genetic epistasis does not inherently require a molecular functional interaction between genes. The formation of a gene family with similar enzymatic function, possibly via a series of duplication events, provides a possible mechanism for the generation of genetic epistasis in the absence of any functional interaction among loci. The simplest model for this form of genetic epistasis is a two-gene system with unlinked loss-of-function alleles, as in the above example of His biosynthesis (see Genetic Subfunctionalization) (Bikard et al., 2009). However, this form of genetic epistasis is not limited to gene families with only two members. It has been predicted that larger gene families whose members are all involved a single biochemical reaction could generate genetic epistasis affecting the related metabolites (Keightley, 1989). This arises because the amount of enzyme does not translate linearly into how much precursor or product is available (Figure 3 ). Instead, enzyme/metabolite relationships typically follow a curvilinear association governed by Michaelis-Menten kinetics. In the example model, an enzymatic activity is encoded by 10 genes, each containing a natural knockout polymorphism that accounts for five units of activity (Figure 3). This model generates a linear gradient of enzymatic activity across the different QTL combinations, which, following Michelis-Menten kinetics, translates into an epistatic surface for metabolite accumulation (Figure 3). This epistatic relationship is not obvious, as most of the single and multiple gene combinations have apparently additive effects upon the metabolite (Figure 3, see range from 40 to 100 enzymatic units). As the number of knockouts increases, however, the organism experiences the enzyme kinetics represented on the left-hand side of the graph, and genetic epistasis can be detected among the loci (Figure 3, see range from 0 to 40 enzymatic units). Thus, theoretically it is possible for the existence of gene families to cause genetic epistasis. This further predicts that epistasis may not affect all trait levels (metabolite, transcript, and physiology) equally (Figure 3, compare enzyme to metabolite).
Gene Family Epistasis.
Michaelis-Menten kinetics controlling the relationship between gene-family polymorphisms and genetic epistasis is shown. The line with diamonds represents the impact of stacking 10 QTL each with two alleles adding a 5 unit (in Km terms) effect upon the accumulation of the enzyme, thus providing a range from 0 units of activity to 100. Michaelis-Menten equations were then used to approximate the amount of product (squares) and substrate (triangles) that would likely accumulate within the plant under this simple model.
Recent metabolic QTL studies have validated the existence of this more complex model of gene family–based genetic epistasis in natural populations. A key step of glucosinolate metabolism in Arabidopsis is the oxidation of a methylthiol to a methylsulfinyl glucosinolate (Kliebenstein et al., 2001c; Hansen et al., 2007). This reaction is controlled by five different genes encoding similar enzymatic activities that also exhibit natural variation in transcript abundance (Kliebenstein et al., 2001c; Hansen et al., 2007; Kliebenstein, 2008; Li et al., 2008). Interestingly, no epistatic influence on transcript accumulation was detected for these loci, although production of the methylsulfinyl product exhibited epistatic interactions (Wentzell et al., 2007; Li et al., 2008) as expected from the model (Figure 3). Further evidence that the existence of gene families can allow for genetic epistasis came from an analysis of expression QTL affecting all identified and predicted biochemical enzymes within Arabidopsis (Kliebenstein, 2008). This study showed that duplicated genes individually have an elevated level of transcript polymorphism due to sequence polymorphisms at or near their physical position. Thus, gene family members are more polymorphic than the average gene and the polymorphisms between individual genes can combine to generate a wide range of phenotypic diversity within the gene family's encoded function. Thus, gene families arising from duplication may be maintained in a genome not only through changes in function but also by the development of complex genetic interactions among gene family members.
Network Architecture and Diffuse Epistasis
A final form of epistasis is diffuse epistasis, and its existence is supported by the image of metabolic QTL provided by metabolic profiling (Whitlock et al., 1995). Diffuse epistasis arises from the control of many biological processes by networks containing complex interactions and multiple crosstalking signals. These highly connected biological networks create more robust links between input and output, protecting biological systems from fluctuations (Barkai and Leibler, 1997; Freeman, 2000; Yi et al., 2000; Dorogovtsev and Mendes, 2002; Spirin and Mirny, 2003; Boccaletti et al., 2006). Although these fluctuations are often considered to be random biological noise, they could also represent genetic variation that is dampened by the network (Figure 4 ). Networks are largely resistant to the effects of single or a few alterations, but they become increasingly likely to fail with the introduction of multiple unlinked defects of small to moderate effect (Albert et al., 2000; Albert and Barabasi, 2002; Agoston et al., 2005; Csermely et al., 2005). Where network structure is known, these failures are to some extent predictable, as shown in Figure 4 for three unlinked genetic mutations that completely disrupt the input-output connection. The corresponding three genetic loci, although not identifying any direct functional epistatic link, would identify a diffuse interaction when viewed in combination. QTL mapping populations contain genotypes that represent a random assortment of hundreds to thousands of independent genetic defects present within the parents and thus serve as an excellent tool to test for multilocus diffuse epistasis (Clark et al., 2007).
Power Network and Diffuse Epistasis.
A simplified network diagram showing the multiple independent paths linking any given input to any related output is shown. The circles can be considered either a gene or the protein encoded by the gene. Lines indicate a mechanistic connection such that information or metabolic flux can pass from one node to another.
(A) The fully functional network with all associated redundancy.
(B) The effects of introducing three polymorphisms (black circles) to the system and the unexpected disruption of the link between input and output.
One impact of diffuse epistasis on plant metabolism is that the metabolic network is vulnerable to combinations of independent mutations. Loss of metabolic network function would then affect other phenotypes, such as growth. This link between metabolite QTL and growth has been observed in several Arabidopsis populations (Keurentjes et al., 2006; Meyer et al., 2007; Lisec et al., 2008; Fu et al., 2009). However, despite predictions that diffuse epistasis occurs in these systems, no significant epistasis was observed in some of these studies (Meyer et al., 2007; Lisec et al., 2008). A potential explanation of this lack of detectable epistatic effects is that diffuse epistasis involves combinations of polymorphisms that would cause extreme phenotypic values, which frequently are treated as statistical outliers and thus eliminated prior to analysis (Lisec et al., 2008). In another experiment querying metabolic QTL, diffuse epistasis appeared to control the tricarboxylic acid cycle (Rowe et al., 2008). Combined polymorphism in at least three loci caused a fourfold to eightfold increase in tricarboxylic acid metabolites as well as stunted plant growth. Only a single specific combination of alleles at the three loci showed a dramatic effect, consistent with expectations for diffuse epistasis (Rowe et al., 2008). Identifying the underlying genetic bases of this observed multilocus effect on metabolism will further test this concept.
Network collapse is most fully explored within the context of power distribution grids, specifically focused on the relation of grid architecture to the frequency of defects that will lead to network collapse (Dobson et al., 2005, 2007; Nedic et al., 2006; Ren and Dobson, 2008). While power distribution grids obviously differ from plant metabolic networks, similarities include the need to balance energy generation with energy use and storage. It may thus be possible to apply power grid distribution theory to our understanding of how complex genetic interactions can generate the plant phenotypic equivalent of network collapse. Power grid distribution theory suggests that the frequency of catastrophic system failures increases with the number of combined defects, with the practical implication that large mapping populations are required to query the potential for these effects (Dobson et al., 2005, 2007; Nedic et al., 2006; Ren and Dobson, 2008). Smaller populations do not allow sufficient sampling of genotypic combinations at more than three loci, causing ascertainment bias in small populations. For instance, in the example in Figure 4, the combination of alleles leading to a network collapse would only be identified in one out of 64 F2 progeny, assuming that this network collapse was not lethal. As such, much larger populations are necessary to study the prevalence and biological impacts of diffuse epistasis.
Application
It has been difficult to incorporate the effects of epistasis into breeding programs due to the increased population sizes required for statistical detection of epistases and the necessity of simultaneously breeding for multiple loci to enhance phenotypes controlled by multilocus interactions. However, recent advances in genomics and genotyping provide the ability to rapidly genotype massive populations, providing the potential to efficiently incorporate epistasis into breeding designs. Functional epistases will be relatively easy to apply to crop improvement once the genes involved in a pathway and their functional interactions are known.
Gene family and diffuse epistasis may be of less immediate practical value and may indeed present an unexpected barrier to some breeding regimes. In the example of gene-family epistasis, to increase the amount of the hypothetical substrate within the crop, a relatively straightforward approach would incorporate loss-of-function mutations at all 10 loci to block the reaction (Figure 4). While this might increase the content of the desired metabolite, it may also cause deleterious consequences if the downstream products have other roles in the plant. As such, gene family and diffuse epistasis raise the potential that the blind stacking of numerous QTL alleles associated with increased accumulation of a given metabolite, without regard for the molecular bases of the QTL and its genetic interactions, could lead to unexpected and possibly unpredictable deleterious consequences. This may occur even if the positive alleles come from different pathways (Figure 4).
Both gene family epistasis and diffuse epistasis represent a breakdown in the relationship between phenotypic consequence and the level of explained phenotypic variance within the population (Keightley, 1989). In this analysis, it was noted that there is not an inherent relationship between the amount of population variance explained by a QTL and the QTL's effect upon the phenotype. This relationship between phenotypic consequence and explained phenotypic variance rapidly breaks down with increases in the involved number of loci or the number of alleles per locus (Keightley, 1989). As such, although it is possible to alter phenotypes dramatically by stacking genes that participate in diffuse and gene-family epistases, large populations are required to first identify these complex higher-order genetic interactions. Additionally, the presence of more than two alleles at any locus in a population will necessitate even more dramatic increases in population size. In the example shown in Figure 4, it would likely take a population of at least 256 F2 progeny to identify just four extreme phenotype individuals; exponential increases in population size could be necessary to provide the statistical support necessary to confidently pursue these results. Most QTL discovery populations are not massive enough for rigorous analysis of the tails of any phenotypic distribution, where genotypes revealing both gene family and diffuse epistasis would reside. Additionally, there is the chance that these genotypes will show a lethal interaction and may only be identifiable by multilocus segregation distortion (Bikard et al., 2009).
CONCLUSIONS AND FUTURE PERSPECTIVES
Recent application of broad-spectrum metabolite profiling to the study of natural variation and quantitative genetics has provided unique insights into the basic underpinnings of plant genetics. These insights may inform numerous fields of research, ranging from gene evolution and network biology to the application of quantitative genetics in crop breeding. As this new information is at present largely limited to two plant species, the expansion of these techniques into other species will test the generality of these observations and reveal the malleability of genetic architecture across phylogenetic distance. Given the intimate link between plant metabolism and physiology, it is likely that the combination of biochemistry and genetics will be as powerful in the future of quantitative genomics as it has proven to be for genetics in the past century.
Acknowledgments
I thank Eva Chan, Heather Rowe, Adam Wentzell, Jason Corwin, and Rachel Kerwin for reading and commenting on drafts of this manuscript as well as for dealing with my ramblings. Funding for this work was provided by National Science Foundation Grant DBI 0820580 to D.J.K.
- Received April 3, 2009.
- Revised May 12, 2009.
- Accepted June 4, 2009.
- Published June 12, 2009.
References
