Heteroblasty refers to the changes in leaf shape and size (allometry) along stems. Although evolutionary changes involving heteroblasty might contribute to leaf diversity, little is known of the extent to which heteroblasty differs between species or how it might relate to other aspects of allometry or other developmental transitions.
Here, we develop a computational model that can quantify differences in leaf allometry between Antirrhinum (snapdragon) species, including variation in heteroblasty. It allows the underlying genes to be mapped in inter-species hybrids, and their effects to be studied in similar genetic backgrounds.
Heteroblasty correlates with overall variation in leaf allometry, so species with smaller, rounder leaves produce their largest leaves earlier in development. This involves genes that affect both characters together and is exaggerated by additional genes with multiplicative effects on leaf size. A further heteroblasty gene also alters leaf spacing, but none affect other developmental transitions, including flowering.
We suggest that differences in heteroblasty have co-evolved with overall leaf shape and size in Antirrhinum because these characters are constrained by common underlying genes. By contrast, heteroblasty is not correlated with other developmental transitions, with the exception of internode length, suggesting independent genetic control and evolution.
The term heteroblasty was coined to describe shoots with abrupt changes in leaf morphology (Goebel, 1900). It is now used more commonly to include the more gradual transitions in leaf shape and size that occur along the shoots of most flowering plant species, even in a constant external environment.
Leaf form is only one of a number of morphological characters that change predictably with plant age and size. Others include the relative positions of leaves on the stem (phyllotaxy), the rate of leaf initiation, differences in epidermal characters such as hairiness, and the ability to flower (reviewed by Poethig, 1990, 2010). Collectively, these changes have been termed phase transitions (Poethig, 1988).
All recognized phase transitions in Arabidopsis respond to a common underlying mechanism. Central to this is a micro-RNA (miR156), which decreases in abundance as plants develop, allowing increased activity of its targets – related SQUAMOSA PROMOTER BINDING PROTEIN-LIKE (SPL) transcription factor genes that overlap in their regulation of different phase characters, including heteroblasty and flowering (Wu & Poethig, 2006; Wang et al., 2008, 2009; Shikata et al., 2009; Usami et al., 2009; Wu et al., 2009; Yamaguchi et al., 2009; Yu et al., 2010). A decrease in miR156 activity, which occurs in response to a signal from existing leaf primordia, has been found to be necessary and sufficient for all the phase transitions examined in Arabidopsis (Wu & Poethig, 2006; Wu et al., 2009; Yang et al., 2011). Parallel studies in a range of species support broad conservation of this mechanism within flowering plants (Chuck et al., 2007a,b; Wang et al., 2011).
The existence of a conserved underlying mechanism raises the question of how different phase transitions might evolve independently of each other. Changes in the relative timing of developmental events – heterochrony – is seen as an important factor in the evolution of morphological diversity in animals (e.g. Gould, 1977), and has sometimes been implicated in plant evolution (reviewed by Li & Johnston, 2000). Because heteroblasty can involve major differences in leaf form within an individual, heterochronic mutations that involve heteroblasty could make a significant contribution to the diversity of leaf form between species (Jones, 1992; McLellan, 1993).
However, very little is known of the extent to which heteroblasty differs between species. This mainly reflects the difficulties involved in quantifying heteroblasty, including the fact that it cannot be described with a simple metric, such as a dimension or ratio, because it can involve differences in both leaf shape and size (allometric variation). Similarly, comparisons between species also have to deal with allometric variation between leaves at different positions in an individual and between individuals. A frequent approach to simplifying this complexity is to assume that some aspects of the variation are more important than others, for example by removing information about leaf size, leaf shape or both (e.g. Harte, 1979; Tsukaya, 2002; Zotz et al., 2011). It is therefore not clear how different aspects of allometric variation between species might relate to one another, for instance how genes that have a similar allometric effect in all leaves might interact with those involved in heteroblasty. The extent to which heteroblasty might vary independently of other phase transitions is also unclear, although its relationship to flowering is of particular interest because flowering time is a life-history trait that can have a major effect on fitness (e.g. Korves et al., 2007).
Measures of allometric variation between leaves and flowers of different Antirrhinum L. (snapdragon) species have previously been made with computational models that captured variation in organ outlines as a limited number of orthogonal principal components, providing a low-dimensional description of much of the difference between species (Langlade et al., 2005; Feng et al., 2009; Wilson & Hudson, 2010).
One advantage of using Antirrhinum to study natural variation in allometry is that the genus consists of c. 25 recognized species that differ in organ shapes and sizes. All are able to form fertile hybrids with each other (Stubbe, 1966; Wilson & Hudson, 2010), allowing allometry genes to be identified as quantitative trait loci (QTLs; Langlade et al., 2005; Feng et al., 2009). Antirrhinum also includes the genetic model Antirrhinum majus L., in which a number of phase transitions have been studied, including variations in leaf size, hair density, phyllotaxy and the ability of plants to flower in response to an inductive photoperiod (Bradley et al., 1996; Cremer et al., 1998; Preston & Hileman, 2010). Antirrhinum majus has also been used to examine the development of heteroblasty, which was found to involve differences in the shapes of leaves at different positions that were established early in development (Harte, 1979; Harte & Meinhard, 1979a,b).
Here we show that computational models can accurately capture different aspects of leaf allometry variation between Antirrhinum species. Heteroblasty is described as the extent to which leaves at different positions on the stem vary independently of each other. The models show that differences in heteroblasty are correlated with variation that involves leaves at all positions in a similar way (overall allometry), so that species with smaller and rounder leaves tend to produce their largest leaves earlier in development. The models also identify variation in heteroblasty that is largely independent of overall allometry. Applying these models to hybrids between species identified similar sources of variation. The resulting measures of allometry allowed underlying genes to be mapped as QTLs and their effects examined in near-isogenic backgrounds. Three QTLs suggested that allometry genes can contribute to variation in heteroblasty in different ways. One gene had a similar effect on leaves at all positions, but could exaggerate heteroblasty because it acted multiplicatively on leaf size. A second acted only in leaves produced later in development, and might therefore contribute to the co-evolution of leaf size and shape and heteroblasty. The third gene altered heteroblasty and internode length, but not other phase transitions, suggesting that heterochronic mutations have contributed to leaf diversity during Antirrhinum evolution.
Materials and Methods
Each species was represented by up to 10 populations covering its geographic range (an average of 8.3 populations per species were analysed). Their origins are detailed in Feng et al. (2009) and seeds are available on request. An F2 population was produced by crossing Antirrhinum charidemi Lange to the inbred line JI.7 of A. majus and self-pollinating a single F1 hybrid. The F2 population (n =204) was genotyped at 156 loci, with an average interval of 4 cM, as described by Feng et al. (2009). Selected F2 hybrids were also self-pollinated for a further six generations and genotyped at locus AmPHB5, c. 5 cM from the most likely position of Q2.4, or Zs016, c. 1 cM from Q2.8, to select heterozygotes. These were then self-pollinated to produce segregating families (n =105 for Q2.4 and n =138 for Q2.8) which were genotyped at the linked markers. Near-isogenic lines (NILs) were produced by back-crossing an F1 to JI.7 for a total of five generations (Rosas et al., 2010). A line heterozygous for parental chromosomes around Q1.3 and Q1.7 was self-pollinated and the progeny (n =236) genotyped for co-dominant polymorphisms in ROSEA (ROS), c. 3 cM from Q1.3 in LG3, and AmANT, c. 3 cM from Q1.7 in LG7. Details of marker loci are given in Schwarz-Sommer et al. (2003, 2010).
To construct allometry models, fully expanded leaves were removed from metamers 2–8 (where cotyledons are part of metamer 1), flattened and scanned, so that each plant was represented by images of seven successive leaves. AAMToolbox (Langlade et al., 2005) was used to position 53 points around the outline of every leaf (Supporting Information Fig. S1) so that each plant was represented by the co-ordinates of 371 outline points. Outlines of leaves from the same metamer in different plants were subjected to a Procrustes alignment, consisting of translation and rotation but without scaling size variation, to superimpose their centroids and minimize variance in the positions of their points. Point co-ordinates for all plants within each population were combined and the remaining variance in their positions partitioned between orthogonal principal components (PCs). Mean leaf outlines for different QTL genotypes in inbred and NIL populations were obtained by a similar Procrustes alignment. To compare leaves from contrasting homozygous marker genotypes in these families, leaf area, length (including the petiole) and width at the widest point orthogonal to the midrib were measured from leaf images using ImageTool (University of Texas Health Science Center, San Antonio, Texas, USA) and tested for significant differences with Student's t-tests.
To detect QTLs, PC values for F2 plants were regressed onto their determined or probable genotypes at 1-cM intervals using QTL Express in a model accounting for both additive and dominance effects (Seaton et al., 2002). The most significant QTL was then fixed as a co-factor for the next round of regression until no further QTLs could be detected above a significance threshold set by permutation of genotypes and phenotypes. The likelihood, position and effect of each QTL were re-estimated in regression with all other significant QTLs fixed as co-factors. Confidence intervals for the position of a QTL were estimated as the regions over which the log-of-odds (LOD) score remained within 1 LOD of its maximum for P =0.99 and 2 LOD for P =0.95.
Heteroblasty models describe leaf allometry within and between plants
To investigate variation in heteroblasty within the genus Antirrhinum, representatives of all its 25 recognized species were grown together in a glasshouse and their fully expanded leaves were flattened and imaged. Points were placed around each leaf outline and the mean position of each point in the data set was plotted to show the average pattern of heteroblasty within the genus (Fig. 1a). Leaf width, length and area all increased progressively from metamer 2 to metamer 5 and then declined from metamer 7 to metamer 8. However, length and width changed disproportionately along the stem, so that leaves from higher metamers had narrower shapes. Heteroblasty in Antirrhinum therefore involves changes in both leaf shape and size (i.e. allometric variation within plants).
The mean leaf shapes were used as a baseline to examine differences between species. Because the positions of the outline points were unlikely to change independently of one another, principal component analysis (PCA) was used on the whole species data set to identify trends in variation. The resulting PCs are ranked according to the proportion of the total variance that each describes and we gave them the subscript ‘spp’ because they refer to differences between species. The variation captured by the first three PCs is shown in Fig. 1(a) as the effect of an increase (red) or decrease (blue) of 2 standard deviations (SD) relative to the mean outline in black. These PCs account for 92% of the variance in the data set, providing a measure of almost all the allometric variation observed.
As PC1spp increases, leaves at all metamers become larger and, because length increases faster than width, they also become narrower in shape (Fig. 1a). PC1spp therefore captures variation that occurs in the same direction at all metamers and involves leaf size and, to a lesser extent, leaf shape. However, the effect of PC1spp is not equivalent in all leaves – those at higher metamers become disproportionately longer as PC1spp increases. PC1spp therefore involves an element of heteroblasty, represented as independent variation between leaves at different positions. This effect is particularly obvious in the shift in the position of the longest leaf as PC1spp changes. Because PC1spp accounts for most (77%) of the variance in the data, it implies that size is the major difference between species and that species with larger leaves tend also to have leaves with narrower shapes and to produce their longest leaves at higher metamers.
While PC1spp captured variation that occurred in the same direction at all positions, an increase in either PC2spp or PC3spp resulted in longer leaves at higher metamers and shorter lower leaves (Fig. 1a). The second two PCs therefore describe heteroblastic variation in leaf length. They also capture heteroblastic variation in leaf width, but with opposite correlations to leaf length in each case. This implies that these characters are weakly correlated in the species, as highly correlated traits should partition into a single PC and uncorrelated traits should each be represented by a separate PC.
Leaf allometry is determined genetically
The variation that is described by the allometry model is likely to have both genetic and nongenetic causes. We assumed that plants recognized as members of the same species were genetically most similar to each other and therefore that the proportion of the total variation that occurred between species provided an estimate of the extent to which each PC was determined genetically. For PC1spp, 95% of the total variance could be attributed to differences between species, suggesting that almost all of the variation is genetically determined. Similarly, 92% of the variance in PC2spp and 84% of the variance in PC3spp could be explained genetically.
We then examined how the allometric variation might relate to the evolution of Antirrhinum. The genus is divided into three taxonomic subsections – Antirrhinum Rothm., Kickxiella Rothm. and Streptosepalum Rothm. – that are largely natural (Rothmaler, 1956; Wilson & Hudson, 2010). Species in subsection Kickxiella were found to share low PC1spp values, while most members of subsection Antirrhinum and Streptosepalum had higher values, reflecting their larger leaves and narrower leaf shapes (Fig. 2a). Most species, though not subsections, also differed significantly from each other along PC2spp or PC3spp (Fig. 2b,c).
From the distributions, we chose two species that occupied different positions along the first three PCs of the allometry model and represented the two major taxonomic subsections. Antirrhinum majus (subsection Antirrhinum) has leaves that are larger and narrower in shape than those of A. charidemi (subsection Kickxiella) and produces its longest leaf at a higher metamer (Fig. 3). These two species occupy opposite extremes of PC1spp and also differ significantly from each other along PC2spp and PC3spp (P ≤0.02 in Student's t-tests; Fig. 2).
To identify genes underlying their allometric differences, we crossed A. majus with A. charidemi and imaged leaves from an F2 population of 137 plants (Fig. 1b). The mean leaf outline for the F2 was similar to the genus, with leaf size reaching a maximum at metamer 6 and leaves at higher metamers having narrower shapes.
PCA showed that the major source of variation in the F2 involved both heteroblasty and overall allometry because leaves at all metamers became larger as PC1F2 increased, with leaves at higher metamers becoming disproportionately longer and narrower in shape. PC1F2 accounted for most (66%) of the variation in the F2 population, revealing that overall leaf size and leaf shape and heteroblasty are strongly correlated in the hybrids, as they are between species. This implies that the correlation is not caused solely by genes involved either in heteroblasty or overall allometry, otherwise it would have been lost due to segregation of the genes in the F2. Rather it suggests that heteroblasty and overall leaf allometry are developmentally constrained by genes that affect both traits together.
By contrast, an increase in PC2F2 involved shorter leaves at metamers 2–5 and longer leaves at metamers 7 and 8 (Fig. 1b). PC2F2 therefore describes the same kind of heteroblastic variation in leaf length as the second two PCs of the species allometry model. However, this leaf length variation is no longer correlated with heteroblastic variation in leaf width in the F2, as width variation is partitioned separately into PC3F2. This suggests that these characters are regulated by different genes and that their correlation in the genus is not the result of a developmental constraint. Although a higher PC2F2 value involved larger leaves in some metamers and smaller leaves in others, it decreased total leaf area slightly.
Mapping leaf allometry QTLs
To further examine the genetic basis for variation along the PCs in the F2 population, we treated each PC as a quantitative trait and mapped the underlying genes as QTLs.
Seven QTLs accounted for 53% of the variance along PC1F2 (Fig. 4). The loci had alleles that all acted in the parental direction – that is, each allele from A. majus increased PC1F2. Most of these loci had been identified previously in a screen for QTLs affecting allometry of metamer 4 leaves alone (Langlade et al., 2005), which is not unexpected given that PC1F2 involves variation in metamer 4 leaves.
Three significant QTLs explained 28% of the variance along PC2F2, which describes mainly heteroblastic variation (Fig. 1b). At all three loci, the A. majus allele increased PC2F2, corresponding to a disproportionate increase in leaf size at higher metamers.
QTLs explaining 72% of the variation in PC3F2 were also detected. Because this PC describes similar changes in width in all leaves (i.e. does not involve heteroblasty; Fig. 1b), its QTLs were not examined further.
Effects of allometry genes in near-isogenic backgrounds
To test further the relationship between heteroblasty and overall leaf allometry, we examined the effects of two major QTLs for PC1F2 (Q1.3 and Q1.7; Fig. 4) in a similar genetic background. We used a population of NILs that had been produced by back-crossing an A. majus × A. charidemi hybrid to its A. majus parent for five generations so that each NIL carried one or more regions of the A. charidemi genome in a genetic background predominantly from A. majus (Rosas et al., 2010). Genome-wide screening with 103 markers identified an NIL that was likely to be heterozygous at Q1.3 and Q1.7, but that carried only A. majus alleles around all the other significant QTLs. This NIL was self-pollinated and its offspring screened with co-dominant markers close to the most likely positions of Q1.3 and Q1.7. The effect of each QTL was then estimated by comparing the mean leaf outline for the two homozygous genotypes (Fig. 5a,b). These estimates are conservative because they do not take into account any recombination between the markers and QTLs that would result in a plant being assigned an incorrect QTL genotype.
Comparing LG7 genotypes confirmed that the A. majus chromosome around Q1.7 increased leaf size at all metamers but did not affect leaf shape significantly (Fig. 5b), consistent with its contribution to variation along PC1F2. Having confirmed segregation of Q1.7 in the NIL, we compared its effect between metamers. One possibility was that the A. majus allele added the same absolute amount to the size of every leaf. Alternatively, it might act multiplicatively with other allometry genes, causing a similar percentage size increase in all leaves. In absolute terms, Q1.7 had it highest effect in metamers 5 and 6, where each copy of the A. majus allele added an average of 128 mm2 (± 16 mm2; SEM) to leaf area, compared with only 97 ± 29 mm2 in higher metamers and 51 ± 16 mm2 lower on the plant. However, when considered in proportion to leaf area, the increases are similar in all metamers – 19% (± 6%) in metamers 5 and 6, 22% (± 6%) at higher metamers and 21% (± 6%) at lower metamers – consistent with a multiplicative effect on leaf size. So, although Q1.7 does not alter the position at which a plant produces its largest leaf, it can exaggerate the heteroblastic differences in absolute leaf size between metamers.
At Q1.3 the A. majus allele significantly increased the lengths of leaves at metamers 5–8, giving them narrower shapes (Fig. 5a). This supported its contribution to the co-variation in overall leaf size and shape that is captured by PC1F2. Q1.3 did not have a significant effect below metamer 5, consistent with a contribution to the independent variation between metamers that forms the heteroblasty component of PC1F2 and so to the constraint between overall leaf size, leaf shape and heteroblasty seen in the species group and F2.
Two major-effect QTLs (Q2.4 and Q2.8) had been found to underlie differences along PC2F2, which describes mainly heteroblastic variation. The effects of these QTLs were examined separately in two inbred families that had been produced from an A. charidemi × A. majus hybrid by seven generations of self-pollination. One had been selected for heterozygosity in a region likely to contain Q2.4 and the other to be heterozygous at markers around Q2.8. Genotypes at a further 241 loci spread across all eight chromosomes detected no other regions of heterozygosity. Each heterozygote was self-pollinated and its progeny genotyped with a co-dominant marker closely linked to its segregating QTL. QTL effects were estimated by comparing the average leaf outlines of homozygotes, as for the NILs. Because the Q2.8 population consistently produced flowers at a lower metamer than Q2.4, presumably because of its different genetic background, only leaves from metamers 2–7 were used to avoid including floral bracts in the analysis.
Plants homozygous for the A. majus marker linked to Q2.4 produced leaves that were significantly shorter at metamers 3–6 and narrower at metamers 4–6 than their homozygous A. charidemi siblings and so tended to reach their maximum length at a higher metamer (Fig. 5c). This supported a role for Q2.4 in heteroblastic variation. The A. majus allele also decreased total leaf area, consistent with the inverse correlation between the metamer with longest leaf and total leaf area captured by PC2F2.
The A. majus chromosome around Q2.8 had no significant effect on leaf allometry (Fig. 5d). One explanation was that alleles of Q2.8 did not segregate, even though the linked marker was heterozygous. Alternatively, Q2.8 might depend for its effect on other genes that were present in the F2 but had been lost from the inbred line.
Effects of leaf allometry loci on other characters
Having identified QTLs underlying inter-species variation in leaf allometry, we tested whether these loci might also influence other aspects of development, including phase transitions such as phyllotaxy, trichome distribution and flowering. These characters were compared between A. majus and A. charidemi and between the QTL genotypes that segregated in the inbred families and NILs produced from them (Table 1).
Table 1. Effects of a leaf allometry locus on other traits
Mean values for the different genotypes at the marker locus linked to the heteroblasty quantitative trait loci (QTLs) in linkage group (LG) 4 are given for Antirrhinum charidemi homozygotes (c/c) and A. majus homozygotes (m/m), ± SEM. Significant trait differences between genotypes are shown at the 0.95 level (*) or 0.99 level (**).
The number of flowers that had opened by the census date provides a proxy for the time taken for each genotype to flower.
Q1.3 and Q1.7 had no detectable effect on any of the other traits that were examined (data not shown). By contrast, plants homozygous for the A. majus allele at Q2.4 were on average 38 mm (± 12 mm) shorter than A. charidemi homozygotes at flowering but produced the same number of internodes (i.e. average internode length was reduced; Table 1). One explanation for this effect is that a single gene at Q2.4 regulates both heteroblasty and internode length, raising the possibility that internode length undergoes a phase transition along with leaf allometry. To examine this possibility, we compared the allometry of successive leaves in an isogenic line with the length of the internodes separating them. Internode length was found to be correlated to leaf length; both increased progressively up the stem to a maximum at metamer 6 (Fig. S2). Therefore, the effect of the A. majus allele at Q2.4 can be interpreted as a shift in characteristics of lower metamers (smaller, rounder leaves separated by shorter internodes) to higher positions on the plant. This is equivalent to slower transitions in two phase characters (heteroblasty and internode length) relative to others, and the effects of Q2.4 can therefore be considered heterochronic.
Here we show that a simple computational model can accurately measure the differences in leaf allometry between Antirrhinum species, capturing almost all of the variation within and between individuals with three principal components (PCs). We show that the same approach identified similar sources of variation in inter-species hybrids, allowing underlying allometry genes to be mapped as QTLs and their effects to be examined in isogenic backgrounds.
The models revealed differences in heteroblasty between plants as the extent to which leaves at different positions vary independently of each other in allometry. They also showed that the main difference between Antirrhinum species (described by PC1spp) involves a correlation between heteroblasty and allometric variation that affects leaves at all positions in a similar way. Subsection Kickxiella species, which have smaller, rounder leaves, therefore produce their largest leaves at lower metamers than species in subsection Antirrhinum. Heteroblasty remained correlated with overall differences in leaf size and shape in the major component of variation (PC1F2) in hybrids between species from subsections Antirrhinum and Kickxiella, implying that these characters are developmentally constrained by common underlying genes. At least one of the QTLs involved in PC1F2 – the locus Q1.3 – might contribute directly to the constraint because in an isogenic background it affected only leaves at higher metamers, where it influences both shape and size. A QTL at this position has a similar effect in petals (Feng et al., 2009), but not other aspects of morphology, suggesting that Q1.3 acts specifically in lateral organs.
In Arabidopsis and maize (Zea mays), changes to the underlying phase information (e.g. miR156 activity) affect all phase transitions and can therefore be distinguished from altered responses, which affect only a subset of transitions (e.g. Cardon et al., 1997; Wu & Poethig, 2006; Wang et al., 2008; Shikata et al., 2009; Yu et al., 2010). Because Q1.3 did not affect multiple phase transitions it seems more likely to be involved in the response to phase information, rather than its creation.
The locus Q1.7 also contributed to the correlated variation in heteroblasty and overall allometry described by PC1F2, but differed from Q1.3 in two respects; it changed leaf size, but not shape, and it affected leaves at all positions. A QTL that affects the size of petals, but not other parts of the plant, also maps to this position, suggesting that Q1.7 is specific to lateral organs (Feng et al., 2009). A further aspect of locus Q1.7 is that it acts multiplicatively with other allometry genes, causing a proportionate size change in all leaves. Such multiplicative interactions are expected of genes with independent effects on the same process, for example that make organs larger by increasing the rate of growth and the duration of growth independently (Sinnott, 1937, 1939). Differences in the shapes of leaves at different positions of A. majus are established early in their development and subsequent growth does not contribute further to these differences (Harte, 1979). Therefore, one possibility is that Q1.7 affects the rate or duration of leaf growth once heteroblastic differences have been established between leaves. Regardless of the mechanisms involved, the multiplicative effects of size genes such as Q1.7 could exaggerate the effects of allometry genes such as Q1.3, contributing indirectly to differences in heteroblasty between species and to the co-evolution of heteroblasty and overall leaf allometry.
The locus Q2.4 was detected for its contribution to heteroblastic variation in F2 hybrids that was largely independent of overall leaf allometry (PC2F2). Its effect was confirmed in an isogenic background, where the A. majus chromosome at Q2.4 slowed the rate at which leaf allometry changes along the stem and decreased internode length. Because both internode length and heteroblasty behave as phase transitions, the effects of Q2.4 are consistent with the action of a single gene. It is not unreasonable to suggest that the allometry of a leaf and its adjacent internode are under common genetic control because they derive from the same group of initial cells within the shoot apical meristem (Jegla & Sussex, 1989).
Unlike the other loci investigated here, Q2.4 had no significant effect on petal development (Feng et al., 2009), as expected of a gene primarily involved in vegetative phase transitions. The effect of Q2.4 can be considered heterochronic, because it changes the timing of two transitions (heteroblastic leaf form and internode length) relative to others. It further suggests that Q2.4 is likely to be involved in the interpretation of underlying phase information, rather than its creation.
The Q1.3 and Q1.7 loci appear to contribute to a developmental constraint involving heteroblasty and overall leaf allometry. The correlated variation in these characters in Antirrhinum species could therefore result from an undirected walk (Feng et al., 2009). Alternatively, selection on one might have contributed to the diversity of both. There is little empirical evidence to support the idea that differences in heteroblasty are adaptive, though this partly reflects the difficulties in separating the effects of different leaf forms within a plant from other factors that change with plant age and size (discussed by Winn, 1999). An alternative explanation is that diversifying selection for overall leaf allometry has given rise to differences in heteroblasty. For example, smaller leaves are considered advantageous when water is limiting (Parkhurst & Loukes, 1972; McDonald et al., 2003), consistent with adaptation of small-leaved Kickxiella species to drier environments such as rock faces (Rothmaler, 1956; Langlade et al., 2005; Wilson & Hudson, 2010). Therefore, selection involving water availability might have driven diversity in both leaf size and heteroblasty between species. The Q2.4 locus involves the opposite correlation between total leaf area and heteroblasty to Q1.3 (the A. majus allele delays the heteroblastic transition and reduces total leaf area) and so contributes towards the independent evolution of these characters. Mutations such as Q2.4 might therefore be favoured if selection for overall leaf allometry has taken either heteroblasty or overall leaf form beyond a fitness optimum. However, Q2.4 has pleiotropic effects on internode length and plant height, and its contribution to differences in heteroblasty might therefore reflect selection on other aspects of plant morphology.
Notably, none of the loci detected for their effects on heteroblasty significantly affected flowering time, even though flowering is a phase transition involving miR156-responsive SPL genes in Arabidopsis (Cardon et al., 1997; Wu & Poethig, 2006; Shikata et al., 2009; Wang et al., 2009), and affects a subset of phase transitions independently (Willmann & Poethig, 2011). Moreover, many Antirrhinum species, including A. majus and A. charidemi, differ significantly in the time and metamer at which they first produce flowers (Table 1; Wilson & Hudson, 2010). This suggests that flowering time, a life-history trait with a potentially large effect on plant fitness, is readily separated from heteroblasty by mutations that allow these characters to evolve independently. This view is consistent with the finding that flowering time can be uncoupled from other phase transitions by environmental cues or mutations in flowering pathway genes in Antirrhinum and other species (Potts et al., 1988; Bradley et al., 1996; Diggle, 1999).
We would like to thank Pat Watson and Bill Adams for growing the plants. This work was supported by BBSRC (grant number BB/D522089/1 and a postgraduate studentship to Y.W.).