Life history in a model system: opening the black box with Arabidopsis thaliana

Abstract

A broad research programme in Arabidopsis thaliana has provided estimates of selection on specific alleles in specific contexts, and identified geographic patterns of alleles in genes linked to timing of flowering. A closely related field has successfully captured many key axes of the evolution of timing of flowering in other monocarpic species through statistical and demographic modelling of large empirical databases. There has as yet been no synthesis between these two fields. Here we examine ways in which the two fields inform each other, and how this synergy will shape our knowledge of life-history evolution as a whole.

INTRODUCTION

Evolutionary theory provides a predictive framework for understanding patterns of variation in natural populations (Sutherland 2005). However, quantifying how evolutionary processes have shaped patterns of traits such as timing of flowering remains a central challenge for evolutionary biologists. The unique body of knowledge on Arabidopsis thaliana has provided novel insights into evolutionary processes. Selection analyses have recently been extended from the lab to transplant experiments designed to directly test adaptation of alleles to natural environments (Weinig et al. 2003). In much of this work, timing of flowering has been a focal trait (Callahan & Pigliucci 2002; Donohue 2002; Korves et al. 2007). In parallel, a body of work addressing optimal timing of flowering in other monocarpic species has sprung up, based on demographic modelling of large datasets (Childs et al. 2003; Metcalf et al. 2003; Rees et al. 2006). So far, there has been little cross-talk between these two fields. The fitness estimates used in Arabidopsis experiments bear little relation to life-history theory and recent developments in the complexities of estimating fitness (Kokko & Lopez-Sepulcre 2007), despite the existence of demographic features in Arabidopsis populations such as density dependence (Thompson 1994; Purves & Law 2002), which can substantially alter fitness outcomes (Mylius & Diekmann 1995). From the other side, the demographic framework used to estimate optimal flowering times remains a ‘black box’ with regard to mechanisms, both for the genetics underlying traits, and the interaction between the environment and physiological processes. This is a fruitful time to consider the linkage between these two fields. Puzzling results from evolutionary demographic modelling of empirical datasets, e.g. the persistence of a range of flowering sizes despite prediction of a single evolutionary optimum, may disappear in a genetically realistic framework. Unifying these two fields provides the means towards a mechanistic framework for moving beyond documenting how climate has shaped flowering allele distributions (Corre et al. 2002) to predicting evolution of flowering time in the face of changing climates.

Here, we start by reviewing what is known about the evolution of timing of flowering and broad areas for development in both fields; we then outline how future research can build from this.

HOW SHOULD FLOWERING TIME EVOLVE?

Research on Arabidopsis

Alleles of genes involved in plastic responses of timing of flowering to the environment such as FRIGIDA (which controls a vernalization requirement for flowering) and PHYTOCHROME C (which alters flowering time in response to day-length) show distinct geographic distribu-tions (Caicedo et al. 2004; Samis et al. 2008). Patterns of polymorphism of these alleles also indicate that they have been subjected to strong selection (Corre 2005; Toomajian et al. 2006) suggesting that these geographic patterns can be at least partly attributed to natural selection. What is less clear is how these selective mechanisms are operating in natural populations (Mitchell-Olds & Schmitt 2006).

Regression approaches have been used to link key fitness components to age or size at flowering (e.g. selection analysis). Resulting selection gradients have been used to predict how the mean value of the trait should change within the population as a result of selection (Lande & Arnold 1983; Rausher 1992), for Arabidopsis grown in growth chambers (e.g. Mitchell-Olds 1996; Scarcelli et al. 2007), glass-houses (e.g. (Clauss & Aarssen 1994; Dorn et al. 2000) and, most recently, in the field (Callahan & Pigliucci 2002; Donohue 2002; Korves et al. 2007). Overall, these analyses consistently conclude that there is selection for increased size at flowering, and generally for decreased age at flowering (Callahan & Pigliucci 2002; Donohue 2002; Scarcelli et al. 2007). Why then are all Arabidopsis populations not flowering at the earliest possible age as large as possible? And how can we connect these results to latitudinal and longitudinal patterns in the field (Caicedo et al. 2004; Samis et al. 2008)?

This selection analysis approach itself may be responsible for the consistency of the estimated directions of selection. Fitness is often measured by taking total numbers of seeds or fruits (Mitchell-Olds 1996; Callahan & Pigliucci 2002; Korves et al. 2007; Scarcelli et al. 2007; Metcalf et al. 2008). Although survivorship may also be considered (Weinig et al. 2003; Korves et al. 2007), the two aspects of fitness (survival and fertility) are rarely combined. Neglecting mortality in fitness estimates can lead to these consistent predictions, as some of the individuals most subjected to selection may never display the phenotype for which selection is being measured (Hadfield 2008). For example, in Arabidopsis, since seed production increases with plant size (e.g. Klinkhamer et al. 1992), the fecundity component of fitness will always indicate natural selection for larger size at flowering (Mitchell-Olds 1996; Callahan & Pigliucci 2002; Donohue 2002; Callahan et al. 2005; Scarcelli et al. 2007). Additionally, in greenhouse studies, even if joint fitness measures are used, unrealistically high probabilities of survival to reproduction will lead to artefactual inference of selection for reproduction at large size. However, older, fecund geno-types should be discounted by their low probability of surviving to such large sizes, allowing intermediate flowering sizes to have high fitness. For an inbreeding species such as Arabidopsis, mortality rates can be incorporated in selection analyses using replicated genotypes, but this has rarely been done.

As fitness is integrated across the life cycle, a major challenge to exploring selection on timing of flowering is that genotypes evolve in concert. Genes ‘for flowering time’ will generally be linked to effects on a range of traits (Loudet et al. 2003; McKay et al. 2003). For example, Korves et al. (2007) showed in an innovative paper that accessions with functional FRI alleles (which lead to later flowering in winter via a vernalization requirement) had higher fitness than accessions with FRI deletions on autumn germination; but lower fitness on spring germination. These fitness differences were associated with higher survival over winter, and smaller seed set in summer (in certain genetic backgrounds). Higher over-winter survival was tentatively attributed to higher water use efficiency (WUE), known to be negatively associated with alleles encoding flowering delays such as functional FRI alleles (McKay et al. 2003). Combining spring and winter results in the context of the life cycle (Box 1) suggests that together these results might reflect operation of a growth–survival trade-off: alleles with higher WUE survive better but grow more slowly, for example, by retaining narrower stomatal apertures (known to mediate a plastic trade-off between growth and survival (Achard et al. 2006). However, this approach also suggests that if WUE increased survival sufficiently, it could be positively rather than negatively associated with flowering delays (Box 1). A joint consideration of the fitness consequences of growth and survival brings to Arabidopsis research, not only prediction of specific directions of the evolution of size at flowering (smaller as well as larger) but also allows predictions about how genes should evolve in concert.

Box 1: Predicting flowering size or age

Fig. 2 indicates how a measure of fitness used in demographic modeling of monocarpic plants (the net reproductive rate R₀) varies as a function of size at flowering, estimated from an Arabidopsis data-set (Supplement). The dashed line illustrates how the optimal flowering size changes if survival conditions worsen. In both cases, R₀ initially rises with size, since larger plants produce more seeds. However, beyond a certain flowering size, R₀ starts to fall, since the chance of surviving to become this large is very low. Eventually R₀ reaches zero for flowering sizes where the probability of reaching these sizes is zero.

Research on other monocarpic species

In other monocarpic species, evolutionary modelling has taken a different approach. Rather than estimating how flowering size should change as a result of selection, the optimal size at flowering is predicted based on empirical data on population demography (Metcalf et al. 2003) using a range of demographic models including fertility and mortality (Kachi & Hirose 1985; Klinkhamer et al. 1996; Rees & Rose 2002; Rose et al. 2002), with extensions to consider the role of environments that change through time (Childs et al. 2004; Rees et al. 2006). This is of particular importance for a timing trait like age or size at flowering, as in a fluctuating environment bet-hedging can become a key advantage associated with reproductive delay (Tuljapurkar 1990).

The emphasis of this body of work has been on using empirical datasets to make predictions about flowering times, based purely on phenotypic data. This has proved successful in a number of cases (Childs et al. 2003, 2004; Rees et al. 2006) but also fails in a number of ways. For example, evolutionary demographic models predict a single evolutionary optimum, even in the presence of fluctuating environments (Childs et al. 2004; Metcalf et al. 2008). Nevertheless, in natural populations of monocarpic species, a broad range of flowering sizes are generally found (Metcalf et al. 2008), presumed to reflect coexistence of different alleles for different flowering sizes given evidence from other populations for heritable variation in threshold flowering size (Wesselingh & Jong 1995; Wesselingh & Klinkhamer 1996).

Arabidopsis information provides a key avenue for reconciling the predictions of no persistence of variation obtained from empirically derived models of monocarp evolutionary ecology with the large body of work in theoretical evolutionary biology that shows how genetic variability around an adaptive optimum can persist (Fisher 1930; Mayo et al. 1990). For example, the empirical models have so far ignored spatial structure (Corre 2005), plasticity of the transition to flowering in response to external and internal cues (Boss et al. 2004; Callahan et al. 2005) (Fig. 1) and genetic architecture. Genetic architecture could facilitate persistence of a range of flowering sizes in two ways. First, the presence of genes of large effect could prevent evolution of the optimal flowering size (Metcalf et al. 2008). Although most genes that affect flowering have small effects (e.g. genes linked to temperature, day length, etc., Lempe et al. 2005), at least a few other loci of large effect are generally identified in surveys of quantitative trait loci (Tonsor et al. 2005), including the vernalization trigger. Although mal-adaptive large effects are likely to be eliminated or mitigated by the effect of many small mutations in large populations, this explanation for persistence of variation in flowering size may be relevant for smaller populations. Second, genetic covariances might equalize fitness of different flowering strategies. Several lines of evidence from Arabidopsis point to higher survival of later flowering accessions, including field studies (Stinchcombe et al. 2004a; Korves et al. 2007), and a genetic correlation between higher WUE and delayed flowering (McKay et al. 2003). If plants with alleles for higher WUE grew slower, but had higher survival, and also flowered later, they might produce the same number of seeds as plants with alleles for low WUE that grew fast, had low survival, and flowered early. Covariation of WUE with alleles for flowering delay consequently has strong implica-tions for coexistence of different flowering strategies (e.g. see Huxman et al. 2008). Finally, alleles controlling flowering size might also be at a mutation–selection balance (Fisher 1930), so that transient deleterious polymorphisms contrib-ute quantitative variation to size at flowering (Mitchell-Olds et al. 2007).

Figure 1.

Theoretical work has shown that within a genetically identical cohort, the optimal flowering time will increase with the quality of the growth environment (Burd et al. 2006), so that if plants could identify their growth environment, they should adjust their flowering time accordingly. The flowering pathways in Arabidopsis are so flexible, with such a diversity of controls and cues, that this is likely (Boss et al. 2004). Here, for individuals of Arabidopsis, Columbia 7 raised in the lab, size at flowering decreased with age at flowering. The fitted model from a linear regression is indicated by a dashed line, taking the form leaf length = 31.4-0.95 age (F_1,39 = 4.2, p < 0.05). This cannot be evidence of escape from the trade-off between size and age at reproduction (Mitchell-Olds 1996), but rather, is likely to represent a plastic shift in the flowering strategy to take advantage of good growing conditions encountered by plants across the experiment, as predicted by theory.

UNITING THESE TWO FIELDS OF KNOWLEDGE

Broadly speaking, Arabidopsis research has linked environ-mental features, such as a cold period or day length, to changes in timing of flowering via the action of specific genes (FRIGIDA, PHYTOCHROME C, etc), and estimated selection on specific allelic variants in particular contexts (e.g. shaded vs. not, summer vs. autumn). By contrast, researchers working on evolutionary demography have ignored the details of environmental context and used the demographic parameters to ask what size or age at flowering would have the highest fitness. What can synergy between these two perspectives bring?

Correlations between flowering time and latitude have been identified in a range of plant species where reproduction is fatal (Reinartz 1984; Lacey 1988; Stinch-combe et al. 2004a), and in Arabidopsis, such geographic patterns have been related to the distribution of allelic variants (Caicedo et al. 2004; Stinchcombe et al. 2004b; Corre 2005; Balasubramanian et al. 2006; Samis et al. 2008). Explanations for these geographic patterns have been proposed at a number of levels. For example, at the purely phenomenological level, Stinchcombe et al. (2004a) pro-posed that as environmental cues for flowering vary systematically with latitude so should alleles associated with responses to the environment. More mechanistically, Toomajian et al. (2006) put forward the (still general) hypotheses that, the recent spread across Europe of FRI deletions, which eliminate the requirement for a cold spell before flowering might correlate with selection by the spread of agriculture for ‘weediness’. Finally, Lacey (1988) identified the specific fitness component responsible for the distribution of flowering size, i.e., lower survival in more southern latitudes selects for faster reproduction.

An exciting goal emerging from the combination of these two fields is the prediction of when alleles should be present at particular locations. For example, we could ask: where should FRI deletions go to fixation, and where should they not? Detailed recording of growth, survival and seed set of local accessions with functional FRI alleles would allow modelling of their fitness (following Box 1 or other methods, Pelletier et al. 2007). Models could then be altered to remove the requirement for a cold spell before flowering, equivalent to a FRI deletion, and the effect on fitness calculated. A recently developed photo-thermal model that successfully predicts timing of flowering conditional on both genetic background and environmental context (Wil-czek et al. 2009) indicates that modelling the effect of an FRI deletion, and thereby predicting evolutionary outcomes for a specific allele is a realistic goal. In some areas, growth may be too slow to allow sufficient seed set to be attained on flowering earlier. In other areas, perhaps the fitness benefits of early flowering are high and we would expect FRI deletions to spread. With sufficient data, complications such as density dependence or variation in growth and survival through time can be accounted for in models (Box 1). The next step would be to go beyond predicting that FRI deletions should spread at a particular geographic location, to identifying what features of the geographic location make it a high fitness location for FRI deletions. The existence of broad consistent patterns in allele distri-butions (Caicedo et al. 2004; Stinchcombe et al. 2004b; Corre 2005; Balasubramanian et al. 2006; Samis et al. 2008) suggest that identifying the key climatic variables is a tractable goal.

Combining this detailed demographic and environmental modelling with reciprocal transplants from local populations would be essential (Mitchell-Olds & Schmitt 2006), as genotypes will tend to evolve in concert and genetic background can affect gene action and fitness (Weinig et al. 2003). For example, FRI deletions might not be able to spread directly in the local genetic background, but could in concert with alleles at other genes favouring earlier germi-nation (Donohue 2002) or particular FLC alleles (Korves et al. 2007). Flowering delays may also covary with WUE (McKay et al. 2008). Consequently, whether rainfall patterns enable or prevent spread of FRI deletions will depend on how FRI deletions covary with WUE and, specifically, how it shapes WUE demography in particular environments.

Moving from spatial variation to temporal variation, global evidence suggests that spring events are occurring c. 2 days earlier (Parmesan & Yohe 2003; Cleland et al. 2007), posing another series of interesting questions. Will plants respond by altering their plastic cues, for example reacting more weakly to signals to grow faster, and flower earlier? Or will genetic variants with lower WUE and earlier reproduc-tion sweep through the population, as selection changes the demographic balance between growth and survival? With a combination of demographic data from wide-spread field populations, and careful modelling of evolutionary out comes informed by known mechanisms such as responses to plant hormones such as ABA and differences between WUE across genotypes, the balance between plastic and static responses in the face of changing environments could be disentangled.

Finally, moving from the ecological to the question of genetic and physiological mechanisms, statistical analysis and experiments for other monocarpic species have suggested that reaching a threshold size is the key requirement before the transition to flowering can take place (reviewed in Klinkhamer 1987; Callahan & Pigliucci 2002). Arabidopsis might also have a threshold size for flowering (which would explain why Arabidopsis in better conditions, where growth is presumably faster, flower earlier; Scarcelli et al. 2007; see Appendix S1). Furthermore, a review across wild Arabidopsis accessions showed less genetic variation in timing of flowering than size at flowering (Lempe et al. 2005). Despite this suggested life-history importance, little is known about the physiological and molecular mechanisms linking size and flowering in Arabidopsis. Size is known to enter into the autonomous pathway that triggers flowering (Boss et al. 2004), and expression of the gene LEAFY has been suggested to accumulate through time (Blazquez et al. 1997) in a manner affected by gibberellin growth hormones (Blazquez & Weigel 2000). Genetic control of the size at flowering is likely to be an interesting area for future research. Although mechanisms are unlikely to be simple, success in untangling epigenetic control of cold repression of flowering (De Lucia et al. 2008) and recent developments in this field (Lister et al. 2008) suggest that whether genetic or epigenetic switches are implicated, their identification is within reach.

TOWARDS A SYNTHESIS OF QUANTITATIVE GENETICS, MOLECULAR PHYSIOLOGY AND DEMOGRAPHIC MODELLING

Researchers studying ultimate causes and consequences are increasingly compelled to reject simplistic, ‘black box’ models, while those studying proximate causes and mechanisms are increasingly obliged to subject their inter-pretations to ecological ‘reality checks’ (Callahan et al. 1997). The timing of flowering in Arabidopsis is ideally suited to both opening the black box and bringing more realistic ecology to a model system, making the prediction of evolutionary changes in timing of flowering in the field a realistic goal. Detailed demographic data on accessions with known alleles, planted into or tracked within natural populations, combined with data on climate will be a key direction for building on recent innovative work (Korves et al. 2007). Although the gap between demographic models (required to define fitness) and detailed physiological models (required to link demog-raphy to the environmental context) remains wide (Tonsor et al. 2005), by matching knowledge of the demographic effects of alleles with different environmental conditions we can sharpen the prediction of outcomes in natural environ-ments, with particular attention to predictions of climate change. Recently, Wilczek et al. (2009) created a statistical model capable of successfully predicting the timing of flowering as a function of genetic background and environ-ment, where the interaction between these two was mediated by the accumulation of ‘photothermal units’. This exciting development suggests that predicting evolution for specific alleles may be within reach.

An immediate step for predicting changes in the flowering time in Arabidopsis is to improve the consideration of mortality. Mortality is a key selection pressure, also likely to be strongest in the smallest (youngest) individuals. Many experiments on Arabidopsis to date involve germinating seeds in the lab and allowing seedlings to grow until they have up to four true leaves before transplanting them into the field (Callahan & Pigliucci 2002), potentially obscuring a key element of the selection process. Obtaining information on this fitness component and including it into estimates of selection pressure should allow quantitative predictions of flowering size beyond the general conclusion that flowering should occur earlier at a larger size.

Beyond this first step, with careful modelling of detailed demographic and environmental data on Arabid-opsis accessions (or other monocarpic species) in a range of natural conditions, linked to common garden experi-ments centred around known genes, quantitative predic-tion of the outcome of selection on flowering size is within our reach (e.g. Pelletier et al. 2007; Metcalf et al. 2008). A final point of note is that evolution occurs via the medium of both selection and genetic drift. Since mutation rates are so well known in Arabidopsis, with more detail on demographic rates within populations, particularly on the magnitude of variance in reproductive success, and connectedness across metapopulations, pre-dicting the spread of deleterious timing of flowering mutations is a realistic goal.

Figure 2.

For individuals of Arabidopsis, Columbia raised in the lab, the net reproductive rate, R₀ for different sizes at flowering, calculated for the demographic models of growth, survival, and flower production defined in Supplementary Table 1 (solid line), using an Integral Projection Model. Methods are outlined in the Supplementary Materials. For an identical model but where the survival intercept (m₀, Table 1) is reduced by 0.8, R₀ is much lower, as fewer individuals survive, and the optimal flowering size is smaller. (The observed flowering size is 2.9 (mm, log scale), i.e. smaller than the optimal, although there is no reason to expect this population to be adapted to the experimental conditions).

The cumulative importance of both survival (to flower at any age, it is necessary to survive to that age) and growth (the advantage of a delay is that there is more time to accumulate more mass that can be invested towards seeds) is coherent with previous work on flowering timing in Arabidopsis. Mitchell-Olds (1996) concluded that plants should indeed flower as early and as large as possible, but were prevented from doing so by a trade-off between being large and young at maturity, which from a dynamic perspective is inevitable, as growth takes time.

This basic framework for fitness estimation has been extended via work on other monocarpic species to quantify the role of number of ecological features including density dependence and variation in demographic rates through time (Rees et al. 2006) on the evolution of flowering time. Both are important. If density dependence is operating, the success of any flowering strategy depends on what other individuals in the population are doing (Kokko & Lopez-Sepulcre 2007). If the environment varies, the evolutionarily stable flowering strategy will change as delaying reproduction is one way in which organisms can hedge their bets.

Applying this set of tools to Arabidopsis has exciting potential. However, several aspects of the ecology of Arabidopsis complicate the accuracy of the types of fitness estimation shown in Fig. 1 including abrupt season endings, i.e., the model in Fig. 1 puts no such upper limit on the time at flowering, but the onset of winter or dry summers may be key for the evolution of timing for flowering in Arabidopsis. The duration of reproduction and growth architecture which are both genetically variable (Schmitt et al. 1995; Schmitt et al. 2003; Scarcelli et al. 2007) complicate measuring fitness of flowering at a certain size or age. Early flowering accessions may have multiple generations within a year (Toomajian et al. 2006). Consequently, the fitness of plasticity responses to seasonal cues may change from one generation to the next, as one generation faces summer conditions and its offspring face winter (maternal effects and epistasis may be one way in which plants adjust to this changing relevance of cues (Galloway & Etterson 2007). Population structure also matters: differences in genetic variability among populations of Arabidopsis suggest patterns of extinctions and re-seeding from founder pools of varying sizes (Corre 2005), which will affect selection (both because fast population growth selects for earlier reproduction, and because selection pressures generally vary across environments (Weinig et al. 2003), but also drift.