Pleiotropic Quantitative Trait Loci Contribute to Population Divergence in Traits Associated With Life-History Variation in Mimulus guttatus
Megan C. Hall, Christopher J. Basten, John H. Willis


Evolutionary biologists seek to understand the genetic basis for multivariate phenotypic divergence. We constructed an F2 mapping population (N = 539) between two distinct populations of Mimulus guttatus. We measured 20 floral, vegetative, and life-history characters on parents and F1 and F2 hybrids in a common garden experiment. We employed multitrait composite interval mapping to determine the number, effect, and degree of pleiotropy in quantitative trait loci (QTL) affecting divergence in floral, vegetative, and life-history characters. We detected 16 QTL affecting floral traits; 7 affecting vegetative traits; and 5 affecting selected floral, vegetative, and life-history traits. Floral and vegetative traits are clearly polygenic. We detected a few major QTL, with all remaining QTL of small effect. Most detected QTL are pleiotropic, implying that the evolutionary shift between these annual and perennial populations is constrained. We also compared the genetic architecture controlling floral trait divergence both within (our intraspecific study) and between species, on the basis of a previously published analysis of M. guttatus and M. nasutus. Eleven of our 16 floral QTL map to approximately the same location in the interspecific map based on shared, collinear markers, implying that there may be a shared genetic basis for floral divergence within and among species of Mimulus.

EVOLUTIONARY biologists have long sought to understand the genetic basis for adaptive divergence between populations with complex multivariate phenotypes. Adaptation to a novel environment may involve evolutionary change of multiple genetically correlated traits as the population approaches a new phenotypic optimum (Fisher 1930; Orr 2000). If variation in individual traits is governed largely by trait-specific loci, the selected traits may be able to evolve independently (unless they are constrained by linkage disequilibrium), whereas those that are governed by pleiotropic loci are going to be evolutionarily constrained. The degree of pleiotropy can have profound effects on the evolutionary trajectory of particular traits (Lande 1979) and therefore on the nature of divergence of multiple traits between populations.

An understanding of the degree of pleiotropy affecting multiple traits sheds light on one of the classic debates in evolutionary biology—whether phenotypic divergence is the result of fixation of one or two mutations of large effect or due to many mutations of small effect. Each of these two options could have different effects on the nature of evolutionary divergence. One of the earliest views of the genetic basis of adaptation was that phenotypic divergence was extremely gradual, consisting of many genes, each having an infinitesimally small effect on the trait (Fisher 1930). This view of adaptation had widespread support among early empiricists (Dobzhansky 1937; Huxley 1942; Muller 1949), although it was later challenged in favor of the alternate view that adaptations were largely the result of substitutions of single genes with large effects (Gould 1980; Gottlieb 1984; Turner 1985). The debate continued (Coyne and Lande 1985; Orr and Coyne 1992) until more recent evidence based on quantitative trait locus (QTL) mapping analyses allowed more rigorous testing of this hypothesis. To date, genetic mapping studies have provided support for both possibilities, some where few large-effect QTL underlie divergence (Doebley and Stec 1991; Bradshaw et al. 1995,1998; Sucena and Stern 2000; Colosimo et al. 2004). Other studies have demonstrated that divergence can result from many QTL of small effect (Liu et al. 1996; Laurie et al. 1997; Zeng et al. 2000; Fishman et al. 2002). However, these patterns are generally interpreted with respect to individual traits rather than accounting for correlations among traits. For example, consider a pair of populations divergent for two traits, and the difference in each trait is controlled by 5 QTL. Phenotypic divergence in these traits could be explained by as many as 10 QTL (if each QTL were completely independent) or as few as 5 QTL (if all QTL were pleiotropic). Because phenotypic divergence between populations generally involves changes in multiple traits for complex organisms, it is particularly important to account for pleiotropic QTL. Examining traits individually to uncover the number of QTL controlling phenotypic divergence could be potentially misleading if pleiotropy exists.

QTL mapping can serve as a powerful way to understand whether multiple traits have diverged together, and it can also be used to distinguish the number and effects of genetic factors controlling divergence in individual traits. Numerous studies have found evidence for pleiotropy affecting multiple traits (True et al. 1997; Jiang et al. 1999; Westerbergh and Doebley 2002; Cui et al. 2004). Of course, fine mapping of pleiotropic QTL can reveal separate, tightly linked QTL (i.e., Knight et al. 2001). Most evidence for pleiotropy is based on overlapping genetic regions detected in separate QTL mapping analyses for individual traits, rather than accounting for the correlated structure of multiple traits in a joint QTL mapping analysis. By joint-mapping QTL affecting multiple divergent traits, we can directly address whether the genetic correlations we see are the result of broadly pleiotropic QTL vs. unlinked QTL affecting separate traits (Jiang and Zeng 1995).

To understand the genetic basis of complex adaptations, we examine the genetic architecture of phenotypic and life-history divergence between two wild, primarily outcrossing populations of Mimulus guttatus (yellow monkeyflower) that differ dramatically in floral, vegetative, and life-history characters. One of the populations we analyze consists of small annual plants that produce thin stems and small flowers (Figure 1; Table 1) and inhabit a high-elevation environment on Iron Mountain, Oregon (IM). These annuals flower rapidly and a large proportion of their meristems are reproductive (i.e., they have flowers or floral buds, rather than leaves). The other population consists of large perennial plants with comparably thicker stems and larger flowers (Figure 1; Table 1) that live in a coastal, temperate environment in the Oregon Dunes National Recreation Area (DUN). These plants flower later than the annuals and a large proportion of their meristems are vegetative (i.e., they have leaves rather than flowers or buds). A reciprocal transplant study reveals evidence for strong local adaptation in each of these two populations (Hall 2005).

Figure 1.—

Parental representatives of M. guttatus populations grown in a common garden greenhouse. Left, M. guttatus from the Oregon Dunes National Recreation Area (DUN); right, M. guttatus from Iron Mountain, Oregon (IM).

View this table:

Phenotypic data for 20 traits measured in M. guttatus on parental lines and hybrids in a common garden

The transition between annual and perennial life histories is common among plants, it is associated with multiple phenotypic differences, and it is the likely response to different ecological conditions (Stebbins 1974). Do life-history characters such as timing of flowering and proportion of floral vs. vegetative meristem growth have a genetic basis? Or are these characters affected only by different environmental conditions? Many life-history characters clearly have a genetic basis (Paterson et al. 1995; Hu et al. 2003; Westerbergh and Doebley 2004). In addition, life-history characters can respond plastically to changing environmental conditions, including the allocation to sexual and vegetative reproduction (Ogden 1974; Schmid and Harper 1985; van Kleunen et al. 2001) and timing of flowering (Alonso-Blanco et al. 1998; Weinig et al. 2002). One of the questions we address in this study is whether life-history divergence between two populations of M. guttatus has a genetic basis or is completely environmentally dependent.

We employ a QTL mapping analysis to understand the genetic basis of divergence in characteristic floral, vegetative, and life-history traits between these two populations. Specifically, we address whether the individual traits differentiating the two populations are affected by many or few QTL and whether these QTL are of large or small effect. We also use multitrait composite interval mapping to clarify whether multiple traits are controlled by pleiotropic (or tightly linked) loci or are controlled by separate genetic loci. Finally, our analysis offers the opportunity to compare the genetic architecture controlling floral trait divergence both within species (our intraspecific study) and between species, on the basis of a previously published analysis that examines many of the same floral traits in an interspecific map of M. guttatus and M. nasutus (Fishman et al. 2002). Floral traits are excellent candidates for investigating the genetics underlying divergence. Floral characters show tremendous variation within and between species, much of which is heritable (Campbell 1996; Galen 1996; Fishman et al. 2002; Hansen et al. 2003). Variation in floral morphology is often the result of adaptive divergence (Grant and Grant 1965; Kim and Rieseberg 1999; Fishman et al. 2002). Here we compare the number, effects, and locations of QTL controlling floral traits and infer whether or not there is a shared genetic basis controlling floral traits between and within species.


Study system:

The M. guttatus species complex (historically Scrophulariaceae, order Lamiales) is highly polymorphic and geographically widespread throughout western North America (Pennell 1951; Vickery 1978; Sweigart and Willis 2003). Populations differ in morphology, mating system, life-history strategy, and habitat type. Although widely studied in ecology and evolutionary biology, taxonomic classification of the M. guttatus species complex has been inconsistent. Some authors have subdivided this taxon into 17 morphologically distinct species (Pennell 1951), while others designate just a few subspecies within the complex (Hitchcock and Cronquist 1973). M. guttatus (2n = 28) is the most common and variable species in the complex.

Populations of M. guttatus can exist as either annuals or perennials, with perennial populations widespread along the Pacific coast. Perennial plants can also be found inland along streams, rivers, and drainage ditches, where there is year-round moisture. Some authors consider the coastal perennial forms, M. guttatus var. grandis Greene, to be distinct varieties from the inland perennials, M. guttatus var. guttatus (Hitchcock and Cronquist 1973). Annual populations are typically located at inland sites like seepy hillside meadows, rocky cliff faces, or road cuts that have abundant soil moisture in the spring and early summer, but little during the late summer. These small annuals are called M. guttatus var. depauperatus (Gray) Grant by some authors (Hitchcock and Cronquist 1973). Plants from these populations are facultative annuals due to seasonally dry environmental conditions, and they can be maintained indefinitely in standard greenhouse conditions. Flower size and vegetative traits differ dramatically between annual and perennial populations, with annuals typically being substantially smaller than perennials for most size-related traits in the field (M. Hall, unpublished results) and in common garden experiments (Figure 1; Table 1). Annual plants flower earlier than perennial plants. Annuals also produce more floral than vegetative meristems compared to perennial plants, which we refer to as proportion of reproductive allocation.

For this analysis, we focus on two populations of M. guttatus that have a high degree of divergence in overall size, habitat, and life history. The well-studied IM population consists of small-flowered, diminutive annuals that live on Iron Mountain, in Oregon's western Cascades (Willis 1993). These plants are predominantly outcrossing (Willis 1993; Sweigart et al. 1999) and have a short period of growth and reproduction, with germination occurring in either the fall or spring, flowering occurring over a 3- to 5-week period in June through early July. All plants at this site die by mid-July. The montane environment experiences fluctuations in temperature and precipitation ranging from below freezing and >6 m of snow in the winter to well above 40° with little or no rainfall in the late summer months (Hall 2005). The DUN population consists of large-flowered perennial plants with larger, nearly succulent leaves that inhabit the temperate environment of Oregon's coastal sand dunes south of Florence in the Oregon Dunes National Recreation Area. At this site, temperatures vary <20° from summer to winter, and there is continual moisture available to plants from heavy rain (up to 2000 mm in the winter months) and coastal fog (Hall 2005). DUN plants typically germinate in the fall and flower from early June through October or November.

Generation of F2 mapping population:

We generated an F2 mapping population from IM and DUN parents to investigate the genetic basis for quantitative trait differences between these populations of M. guttatus. The IM parent is a highly fertile inbred line (IM62) derived from the Iron Mountain site. This parental line is the same parental line used to construct the previous interspecific map (Fishman et al. 2001). Two separate wild-collected plants (DUN1 and DUN2) were used as parents from the DUN perennial population. Each of the DUN parents was reciprocally crossed to IM62 to produce four sets of F1 individuals, and one plant from each class was selected at random to produce the F2 generation. One F1 plant (IM62 maternal parent, DUN1 paternal parent) was reciprocally crossed to another F1 plant (DUN2 maternal parent, IM62 paternal parent) to produce two sets of F2 seeds. The other two F1 plants were also reciprocally crossed to each other to produce two other sets of F2 seeds, for a grand total of four sets of F2 seeds. Each F2 individual therefore has a nuclear genome derived from contributions of three individuals (IM62, DUN1, and DUN2) and a cytoplasmic genome derived from either the DUN or the IM population. Note that this crossing design enforces outbreeding with respect to alleles derived from the DUN population, but allows for homozygosity of alleles from the highly viable and fertile highly inbred IM62 line, thereby reducing the potential for transmission ratio distortion in F2 progeny to be caused by inbreeding depression. All seeds used in the common garden experiment described below were the same age: the F1 plants and the parental plants were recreated by selfing IM62 and reciprocally crossing DUN1 and DUN2 at the same time as the creation of the F2 lines.

In June 2000, we grew 100 IM62 plants, 50 each of the DUN1 × DUN2 plants and their reciprocal crosses (see above), and 200 F1 plants along with the F2 mapping population (N = 600 total, with each of the four F2 classes equally represented) in individual pots in a common garden experiment at the University of Oregon Department of Biology greenhouse. Plants were grown in 4-in. pots filled with sand over a thin layer of hemlock bark on the bottom, to prevent sand from escaping the pot. A thin layer of organic potting mix (Black Gold potting soil; Sun Gro Horticulture, Bellevue, WA) was sprinkled on top to prevent seed dessication. We planted five seeds of the same class per pot on June 12, 2000, and pots were placed in flats in a fully randomized design in the greenhouse during the long days when flowering begins for each of the native populations. Plants were watered as needed two to three times daily and left unfertilized. Germination rates were measured per pot, and seedlings were thinned to the centermost individual after germination, 2 weeks after planting.

Phenotypic analyses:

We measured 20 floral, vegetative, and life-history traits on all plants that flowered, using an engineering ruler with gradations to the nearest 0.01 inch. All measurements were converted into millimeters. To estimate overall plant size (vegetative characters), we measured the length, width, and thickness of the first two leaves on each plant at the time of its first flower. At this time, we also measured stem thickness at the base of the plant (between the first true leaves and the cotyledons) and the internode length between the first and second set of true leaves. Stem and leaf thickness were measured with digital calipers to the nearest 0.01 mm. If the vegetative traits continued to grow after flowering, there could be an association of these traits with flowering time. However, we chose to measure traits for all plants at a defined developmental stage. We also recorded the date of flowering for the first two flowers per plant and used the average of these 2 days, and for each of these flowers we measured six floral size traits (corolla width, corolla length, corolla tube length, style length, stamen length, and distance separating stigma and nearest anther). For a diagram of these floral traits, see Fishman et al. (2002).

In addition to date of first flowering, we measured a number of other life-history traits. After 10 weeks from planting, we counted the total number of floral and vegetative meristems on each plant. Floral meristems were scored as any stem bearing flowers or flower buds, and vegetative meristems were lacking any flowers or flower buds. The percentage of reproductive allocation was estimated for each plant by dividing the number of floral meristems by the total number of meristems (floral plus vegetative). After 16 weeks, the soil and sand were washed from the roots of each plant and the plants were placed in labeled paper bags. Each bag was placed in a drying oven for 3 days on lowest heat to remove all of the moisture from the plants. Dried plants were then weighed on an electronic balance to the nearest 0.1 g with their roots (total mass) and with the roots removed just below the cotyledons (above-ground mass). The soil granules were nearly impossible to remove from the roots, particularly for the annual plants; therefore the total mass was not included in the data set.

We measured male fertility by collecting the anthers from the first two flowers on each plant and placing them in 60 μl of lactophenol aniline blue stain (Kearns and Inouye 1993). We counted the number of viable (darkly stained) and inviable pollen grains in a 0.8-μl subsample of each collection under a compound microscope. The aniline blue dye stains intact (starch-filled) cytoplasm, which may also be present in some inviable grains; therefore our estimates of pollen fertility may be slightly relaxed. Total number of pollen grains was also calculated as a summation of viable and inviable pollen grains. For each plant, we divided the total number of viable pollen grains by the total number of pollen grains measured to estimate the percentage of viable pollen per flower.

For each trait measured, we calculated the mean and variance for each class (IM parent, DUN parent, F1, and F2) and for each of the F2 classes separately. The F1 hybrids are mostly genetically homogeneous, so the phenotypic variance of this class reflects just environmental variance, whereas the F2 phenotypic variance reflects both environmental variance and the segregation of alleles at genetic loci differentiating the parental lines. For the environmental variance (VE), we used the F1 phenotypic variance. We also estimated the average variance within F1 classes to account for any differences among classes. These estimates were very similar to, although slightly smaller than, on average, the VE calculated from the F1 phenotypic variance; therefore we simply used the latter. The environmental standard deviation (ESD) for each trait was calculated as the square root of VE. We calculated the genotypic variance as VG = Var(F2) − VE and then estimated the broad-sense heritability for each trait as H2 = VG/Var(F2). Genotypic correlations were estimated by calculating covE from the F1 class and for each pair of traits and then estimating covG = cov[F2] − covE. Genetic correlations (rG) among traits were calculated as covG(i, j)/sisj, where covG(i, j) is the genetic covariance between traits i and j and si and sj are the square roots of the genotypic variances of the two traits, respectively. Genetic correlations were not calculated for traits with negative estimates of H2. Phenotypic correlations were also calculated among traits for the F2 hybrids.

We performed a simple statistical test for epistasis (Lynch and Walsh 1998), using analysis of variance (ANOVA) to calculate the class means and sampling variances for each trait. These were used to calculate the test statistic,Math(1)where z is the trait mean for each class. In the absence of epistasis, Δ is expected to be zero. The ratio Mathis a t-test for epistasis or a rejection of a purely additive-dominance model. In addition, the ratio of Δ to the F2 mean expected under the additive-dominance model (E[F2]) is a relative measure of the severity of hybrid breakdown.

Linkage map construction:

In a previous analysis, we constructed a linkage map for this F2 population (N = 539) at 154 AFLP, microsatellite, and gene-based markers (Hall and Willis 2005). The linkage map spans 1482 cM Kosambi, includes 14 linkage groups (which presumably correspond to the 14 pairs of chromosomes in M. guttatus), and has an average interval length of 15 cM. We detected transmission ratio distortion in nearly half of all markers. Our most distorted marker, LFY, had a normal percentage of DUN homozygotes, excess numbers of IM homozygotes (216 observed vs. 119 expected), and fewer than expected heterozygotes (143 observed vs. 238 expected). Although the presence of distorted markers diminished some of our power to detect QTL, distortion was not so severe as to eliminate entire genotypic classes; therefore we feel that it had a minimal impact in this study.

Quantitative trait locus analyses:

We mapped QTL for 20 single traits using composite interval mapping (CIM; Zeng 1993, 1994) and for subsets of traits using multitrait composite interval mapping (MCIM; Jiang and Zeng 1995), using QTL Cartographer v. 1.17 (Basten et al. 2002) and QTL Cartographer Windows 2.0 (Wang et al. 2005). For each trait, the CIM procedure tested the hypothesis that a test site in an interval between adjacent markers had a QTL affecting the trait, while accounting for genetic background by using multiple regression on additional markers as cofactors. The cofactors included in each CIM model were determined by forward–backward stepwise regression, with the critical P-values set at 0.05. Tests were performed at 2-cM intervals with a flanking window size of 10 cM. The likelihood-ratio (LR) test statistic is −2 ln(L0/L1), where L0/L1 is the ratio of the likelihood under the null hypothesis (there is no QTL at the test site) to the alternative hypothesis (there is a QTL at the test site). Experimentwise significant levels (α = 0.05) were determined by permuting the phenotypes against the genotypes 1000 times for each trait (Churchill and Doerge 1994).

Because many of the traits (particularly the floral and vegetative characters) were highly correlated and the single-trait CIM analyses identified QTL for multiple traits mapping to the same interval, MCIM was used to jointly map QTL affecting (a) six floral traits (corolla width, corolla length, corolla tube length, stamen length, and style length), (b) four vegetative traits (stem thickness, leaf width, internode length, and leaf thickness), and (c) six general traits [including representative floral (corolla width, corolla tube length), vegetative (stem thickness, leaf width), and life-history traits (days to flower, percentage of reproductive allocation)]. These traits were chosen on the basis of their relatively high heritabilities and genetic correlations. The MCIM procedure is similar to single-trait CIM, but the LR test statistic is −2 ln(L0/La), where La is the likelihood under the alternative hypothesis that the test site is a QTL affecting any of the included traits. MCIM provides additional power and accuracy for mapping QTL by taking into account the correlational structure of the phenotypic data (Jiang and Zeng 1995). Experimentwise significance levels (α = 0.05) were determined by permuting the phenotypes against the genotypes 1000 times so that the correlations between traits were maintained (Churchill and Doerge 1994).

To determine if QTL detected by MCIM had pleiotropic effects on the traits in each analysis, individual MCIM likelihood-ratio test values were examined for each position where joint mapping indicated the presence of a QTL (Jiang and Zeng 1995). Pleiotropy was indicated by the rejection of the null hypothesis of no more than one trait having a LR test value greater than a significance threshold value of 5.99 (Math) at a particular QTL position as determined by the model parameters estimated jointly by MCIM. This test does not require corrections for multiple tests along the genome because each position is fixed prior to the test, which increases the power to detect QTL effects on multiple traits (Jiang and Zeng 1995).


Phenotypic analyses:

Plants from the two M. guttatus populations grown in a common garden were highly divergent for many of the traits measured, indicating that these differences have a genetic basis (Table 1). For floral characters, the parental lines differed by 4–11 ESDs, and the mean of the DUN plants was greater than that of the IM line. The parental lines differed by 2–7 ESDs for vegetative traits. The DUN plants were larger for all vegetative characters, with the exception of internode distance. All of the floral and vegetative traits appeared to be additive, with the F1 and F2 means nearly intermediate between the two parental means. Broad-sense heritabilities (H2) were small to moderate for the floral and vegetative characters (0.22–0.65; Table 1). The variance of the DUN parents was typically larger than that of either the IM parents or the two hybrid classes, consistent with the variance scaling with the mean.

Life-history and male fertility traits showed less difference compared with floral and vegetative traits between parental lines (0.4–3 ESDs) and had primarily low heritabilities (0.031–0.61; Table 1). The IM plants flowered earlier, produced many floral meristems and few vegetative meristems, had reduced mass, and produced less total pollen on average than the DUN parental lines. The life-history and fertility traits do not appear to be entirely additive. For example, for flowering date, both F1 and F2 hybrid classes flowered early like the IM parent, which is consistent with partial dominance toward the IM parent. However, this trait also deviated significantly from the predictions of an additive-dominance model of inheritance (Table 1), suggesting some level of epistasis controlling flowering time. All of the meristem traits show partial dominance in the hybrids toward the parent with more meristems (IM for floral meristems and DUN for vegetative meristems). For male fertility traits, the DUN plants made more pollen grains (both viable and inviable) than the IM parents (Table 1), although the fraction of viable pollen produced did not markedly differ between the parents or the hybrid classes. The F1 hybrids produced more viable and less inviable pollen than expected under a strictly additive model, while the F2 hybrids had the opposite pattern—they made less viable pollen grains, more inviable pollen grains, and less total pollen than expected. All of the pollen grain measures (with the exception of fraction of viable pollen) were inconsistent with the additive-dominance model (Table 1), which implies that epistatic interactions are involved in control of pollen production.

Genetic and phenotypic correlations:

All of the floral size traits, with the exception of stigma–anther separation (SA), were strongly and positively correlated with each other, both genetically and phenotypically (Table 2). The total number of pollen grains produced (TP) was also highly positively correlated both genetically and phenotypically with the floral size traits. The vegetative traits had weak to moderate positive genetic correlations with each other, although internode length was negatively correlated with other vegetative traits. Phenotypic correlations between vegetative traits were moderate to high, with the exception of internode length. Life-history traits were not strongly genetically correlated with each other, although flowering date and percentage of reproductive allocation had modest negative genetic and phenotypic correlations with each other, and they represent important indicators differentiating our annual and perennial populations. Flowering time had a strong positive genetic correlation with both corolla width and stem thickness, and floral and vegetative traits were highly correlated (both genetically and phenotypically) in a positive direction. Due to the nature of the genetic correlations, we chose to group our traits in three sets for further investigation of the cause of genetic correlations among multiple traits. First, we grouped six floral traits; then we grouped four vegetative traits; and finally we grouped six representative floral, vegetative, and life-history traits to determine the extent of pleiotropic QTL affecting multiple traits.

View this table:

Genotypic (above diagonal) and phenotypic (below diagonal) correlations in intraspecific F2 hybrids

Quantitative trait locus analyses:

Floral QTL:

We identified 16 putative QTL affecting one or more floral traits on the basis of the LR statistic profile of the joint MCIM model (Figure 2A). Twelve LR peaks exceeded the threshold of 43.52 (estimated by permutations, α = 0.05), although we accepted 5 lower peaks that all had highly significant LR profiles for one or more floral traits on the basis of the single-trait analysis produced by CIM and MCIM (data not shown). These peaks are located on linkage group (LG)4, LG6, LG8, LG11, and LG12. One significant peak on LG1 did not affect any of the six floral traits and was therefore not included.

Figure 2.—

Likelihood-ratio (LR) test statistic profile from multitrait composite interval mapping of: (A) six floral traits; (B) four vegetative traits; and (C) six representative floral, vegetative, and life-history traits in the intraspecific F2 individuals of M. guttatus. The solid line indicates the LR significance threshold for joint mapping generated by permutation analyses (α = 0.05, experimentwide). LR thresholds are 43.52, 40.18, and 55.97 for floral, vegetative, and multiple-traits analyses, respectively. All linkage groups are along the x-axis, with vertical double lines separating them. The positions of mapped markers (Δ) are shown along each linkage group. QTL detected were labeled numerically and with a subscript (floral, vegetative, or multiple traits) for each separate analysis (see text for explanation of individual peaks).

All of the individual floral traits were polygenic (mean number of detected QTL per trait = 8.7, range = 5–12). The direction of allelic effects was consistent, in general, with the phenotypic differences between parents (i.e., the larger-flowered DUN carried the “plus” allele). All 7 QTL affecting stamen length were positive, and most other floral traits had the majority of QTL in the positive direction with just a few negative QTL (i.e., 8 of 12 QTL affecting corolla width were positive). Many of the QTL showed partial dominance of one parental allele, but there was no overall pattern of directional dominance (Table 3A). One QTL (QTL10f) appeared to be overdominant and another QTL (QTL11f) appeared to be underdominant. These results could reflect true overdominance or underdominance, but they could also result from a low density of markers (particularly codominant markers) in these regions that makes it difficult to detect QTL tightly linked in repulsion.

View this table:

QTL for floral traits from LR profile (Figure 2)

We used two different methods to estimate the magnitude of effects of individual QTL on each trait. One biologically relevant measure of QTL size is to scale the effect of substituting a single QTL by the difference between populations. Here, we standardized 2a by the difference in the parental means (Table 3B). Using this method, the floral QTL we detected had a range of individual effects from very small (QTL 11f, <1% of the parents' difference in corolla tube length) to large (Table 3B, QTL 5f, 24% of the populational difference in stamen length). Each floral trait had at least one QTL that explained >17% of the species difference, but the remaining QTL were small. QTL 5f on LG8 had a consistently large effect on multiple traits (Table 3B). We also estimated the magnitude of QTL effects relative to the ESD (Table 3B). This method reveals that most QTL have small effects, although several have larger effects. In 45 of 52 of the QTL/trait combinations, the substitution of one parental genotype for the other caused a change in phenotype equivalent <1 ESD. The remaining 7 QTL caused a change of >1 ESD for individual traits, most of which is attributed to the effect of QTL 5f, which affected all six floral traits, four of which had homozygous affects >1 ESD.

We used Jiang and Zeng's (1995) test for pleiotropy to determine which traits each floral QTL affected. Nearly all of the floral QTL (14 of the 16) identified by MCIM had significant effects on multiple traits (Table 3). One of the exceptions identified only affected style length (QTL 12f), and the other affected only total pollen production (QTL 1f).

Vegetative QTL:

We identified seven putative QTL affecting one or more vegetative traits on the basis of the LR statistic profile of the joint MCIM model (Figure 2B). Four LR peaks exceeded the threshold of 40.18 (estimated by 1000 permutations), although we also accepted three marginally significant peaks (on LG6, LG7, and LG11) on the basis of single-trait LR profiles produced by CIM and MCIM (data not shown). As with the floral traits, the direction of QTL effects on vegetative traits was consistent with the phenotypic differences between parents. No overall pattern of directional dominance was obvious, nor was there evidence for overdominance (Table 4A). We detected fewer QTL for vegetative traits than for floral traits (mean number of detected QTL per trait = 4.3, range = 3–5; Table 4A).

View this table:

QTL for vegetative traits from LR profile (Figure 2)

Our two methods of estimating QTL effect on vegetative traits produced similar results to those of the floral traits analysis. Using the method where we standardized the additive effect of a QTL by the difference in parental means, we detected QTL with a broad range of effects from very small (QTL 1v, 7.7% of the parental difference in stem thickness) to very large (Table 4B, QTL 7v, 91.2% of the parental difference in internode length). This same very large QTL also had a pronounced effect when we analyzed the difference with respect to the ESD, with a substitution of one parental genotype for the other causing a phenotypic change equivalent to 1.4 ESDs. Although the extent of the effect is difficult to define, in both cases this QTL appears to be sizeable. Overall, when the additive effect of a QTL is scaled relative the ESD, we found that 6 of 17 vegetative QTL had effects >1 ESD, while the remaining QTL had small effects.

Six of the seven vegetative QTL detected affected multiple traits using Jiang and Zeng's (1995) test for pleiotropy (Table 4). One exception affected only leaf thickness (QTL 3v).

Multiple-trait QTL:

Five QTL were identified that affected one or more of our six representative floral, vegetative, and life-history traits on the basis of the LR statistic profile of the joint MCIM model (Figure 2C). These five LR peaks exceeded the permutation threshold of 55.97.

We found that relatively few QTL explained the trait differences between parents (mean number of detected QTL per trait = 2.5, range = 2–4). The direction of QTL effects was generally consistent with the phenotypic differences between parents (Table 5A), and there was no pattern of directional dominance. By scaling the additive effect of a QTL by the difference in parental means, we detected QTL with a broad range of effects from very small (QTL 5m, <1.0% of the parental difference in corolla tube length) to large (Figure 3, Table 5B, QTL 5m, 36.1% of the parental difference in leaf width). Most of these individual QTL effects were moderate, explaining between 10 and 20% of the phenotypic differences between parents. Alternatively, when scaling each QTL's individual additive effect by the ESD, 5 of 15 QTL had homozygous effects >1 ESD, which we consider to be moderate to large QTL.

Figure 3.—

Additive effect and direction of effect of major QTL (on LG8, marker 3) on: (A) corolla tube length and stamen length from the floral traits analysis (QTL 5f), (B) corolla tube length and stem thickness from the multitrait analysis (QTL 3m), and (C) corolla tube length and leaf width from the multitrait analysis (QTL 3m). Parental means for each trait are plotted with bars indicating standard deviation. QTL additive effect is positioned at the midparent with the homozygous effect of substitution indicated for each trait.

View this table:

QTL for multiple traits from LR profile (Figure 2)

All five QTL affected multiple traits (Jiang and Zeng 1995, Table 5). This analysis confirmed that two of the major QTL detected separately in the floral and vegetative traits analyses (both on LG8) affect both floral and vegetative traits. These two QTL also affect life-history traits (QTL 3m affects the percentage of reproductive allocation, QTL 4m affects days to flowering).


The two populations of M. guttatus studied in this common garden experiment differed markedly in many phenotypic traits associated with life history and morphology, indicating a genetic basis for the divergence. As expected from observations of the phenotypes in nature, the annual plants from Iron Mountain in Oregon's western Cascades (IM) had smaller flowers and vegetative traits, flowered earlier, and produced more floral meristems relative to vegetative meristems on average than the perennial plants from the coastal Oregon sand dunes (DUN). Our investigation of the genetic basis for floral, vegetative, and life-history divergence revealed substantial numbers of pleiotropic quantitative trait loci governing complex phenotypic divergence and also indicated that these classes of traits have different genetic architectures. Overall, all of the traits were controlled by at least two QTL, and we detected several large-effect QTL.

Number of quantitative trait loci:

The divergence in floral and vegetative traits between populations of M. guttatus is controlled by many QTL. Despite the large number of QTL detected, the sum of QTL effects for each floral and vegetative trait is <75% of the difference between parents. The remaining unexplained difference suggests that there are many QTL that were not detected in our study. If true, then the divergence involves a much larger number of genes controlling phenotypic divergence. Alternatively, epistatic interactions among detected QTL may be responsible for the unexplained difference. Because each QTL may contain multiple linked genes, and methods of estimating gene number are inherently biased toward underdetection of QTL and overestimation of QTL effect (Beavis 1994; Zeng 1994), the 16 floral QTL and 7 vegetative QTL detected in this study are minimum estimates of gene number. In addition, each gene could contain multiple substitutions that affect different traits.

When we combined a subset of the floral and vegetative traits with two life-history traits in a multitrait analysis, we detected fewer QTL overall and fewer affecting each trait. This contradicts our previous results for four of the six traits. For example, we detected only 3 QTL affecting corolla width using multiple-trait analysis, compared to 12 QTL in the floral traits analysis, where we concluded that this is likely to be a minimum estimate of gene number. In this third multitrait analysis, the lower genetic correlations among traits (particularly the life-history traits) may inhibit our power to detect QTL of smaller individual effect on certain traits. While we are certain that floral and vegetative traits are governed by many loci, the number of QTL controlling life-history traits is less clear, although our results suggest the number is small.

Our results of mostly polygenic trait divergence between populations are consistent with other studies among different accessions of Arabidopsis, where floral, vegetative, and life-history traits are mostly polygenic, with 2–15 QTL detected per trait (Mitchell-Olds 1996; Alonso-Blanco et al. 1998; Juenger et al. 2000; Perez-Perez et al. 2002; Ungerer et al. 2002). Unfortunately, no consistent patterns emerge to explain why certain traits are more or less polygenic. One possibility is that different traits may have experienced different patterns of selection, which could affect the numbers of QTL responsible for trait divergence. In a review of the literature where selection differentials were measured, Kingsolver et al. (2001) found that the strength of directional selection differed between morphological and life-history traits, where selection was generally stronger on morphology. We need more studies that investigate both the genetic basis controlling trait divergence and the ecological significance of particular traits in the wild to better understand how different patterns of selection might affect the total number of genes controlling adaptive trait divergence.

Comparative mapping of floral QTL within and between Mimulus species:

To what extent is the genetic architecture controlling floral divergence shared within and between species? Comparative mapping of the same traits can reveal whether there are shared QTL locations and numbers within and beween species. In this interpopulational study we measured five of the same floral traits as those studied in an interspecific QTL mapping study of M. guttatus and closely related self-fertilizing M. nasutus (Fishman et al. 2002). These two studies have 27 markers in common (Hall and Willis 2005). To understand the extent to which there were shared QTL affecting floral divergence between and within species of Mimulus, we made two comparisons. First, we compared the total number of QTL detected in each study. We found a total of 16 floral QTL compared to the 24 detected interspecific floral QTL, lending support for the hypothesis that QTL number is positively correlated with genetic divergence (Kim and Rieseberg 1999). However, the smaller number of QTL detected in this study may also be caused by slightly reduced statistical power due to the smaller number of codominant markers.

Unfortunately, there are very few existing systems where QTL analyses have examined divergence between both populations and species, particularly for comparable traits. QTL affecting grain weight were mapped in both intraspecific (Yu et al. 1997; Xing et al. 2002) and interspecific crosses (Moncada et al. 2001; Li et al. 2004) of rice, where more QTL for grain weight were detected in the intraspecific crosses relative to the number detected between species. However, it is difficult to directly compare patterns of phenotypic divergence between any pair of studies (including Mimulus), as they were not conducted in the same environment. Furthermore, domesticated and wild systems have experienced very different evolutionary histories; therefore direct comparisons between the two may be limited. Clearly, we need more studies that compare the genetic basis for phenotypic divergence both at the intra- and at the interspecific level to understand whether a pattern exists between the degree of genetic divergence and QTL number.

Second, when we compared the locations of floral QTL between both maps, we found that 10–11 QTL map to approximately the same locations (Figure 4). These shared QTL suggest the possibility that some of the same underlying genes could be responsible for divergence in floral traits between and within species of Mimulus. One shared QTL involves QTL 5f and the interspecific QTL 13, both tightly linked to marker CYCB on LG8. Interestingly, these floral QTL actually affect different sets of traits in the two studies. If these QTL are caused by the same genes, then they seem to have very different effects on floral traits within vs. between species of Mimulus. Of course, each QTL spans a fairly broad genomic region that may contain hundreds of genes, and the colocalization of the QTL may simply be due to chance. Fine-scale mapping with additional markers and ultimately, positional cloning, may help distinguish whether some of these “shared” QTL affecting floral divergence are truly controlled by the same underlying genes. However, some QTL clearly mapped to different locations in the two maps, indicating that floral divergence may be evolutionarily labile with multiple alternative genetic changes involved in different lineages.

Figure 4.—

Comparative map of floral QTL within and between species of M. guttatus and close relative M. nasutus. The linkage group is indicated above both intraspecific (g × g, M. guttatus × M. guttatus) and interspecific (g × n, M. guttatus × M. nasutus; Fishman et al. 2001) maps. Hatch marks indicate marker placement. Only terminal markers and common markers are labeled on each map, with thin lines connecting markers in common. For more detailed description of map comparisons, see Hall and Willis (2005). LGs with single common markers are matched up arbitrarily; note that the orientation could be rotated. Arrows point to the location of QTL affecting one or more floral traits. Numbers alongside arrows correspond to the QTL number given (see Table 3 for intraspecific numbers and Fishman et al. 2002 for interspecific numbers). Shaded solid bars are putative shared floral QTL between the two maps. Shaded gradient bars (LG14) are alternative putative shared floral QTL, depending on orientation of LGs.

In rice, there is evidence for shared QTL affecting grain weight in either intra- or interspecific crosses (Yu et al. 1997; Moncada et al. 2001; Xing et al. 2002; Li et al. 2004). In one interspecific study, fine-scale mapping demonstrated that one of the potentially shared QTL regions affecting grain weight in rice contained 14 genes (Li et al. 2004). There are currently no similar studies in intraspecific rice that determine whether any of these genes are actually shared between and within species. Clearly, this is an important avenue for further research.

Effects of quantitative trait loci:

There is much interest in understanding whether adaptive divergence is due to major or minor genes (Orr and Coyne 1992). However, there is no standard criterion for defining major vs. minor QTL. Furthermore, QTL effect sizes can differ dramatically depending on how they are estimated (Lexer et al. 2005). Tanksley (1993) characterized QTL as potentially major if they explained >10% of the phenotypic variation in the segregating population, generally referred to as percentage of variance explained (PVE). This is the most typical measure used to estimate QTL effect, and it may be particularly appropriate for lab or agricultural systems. However, a more useful measure for understanding adaptive divergence in the wild may be to estimate QTL effect in terms of the difference between parental populations or relative to the phenotypic variation within populations. For example, True et al. (1997) uses a fairly stringent criterion by defining a major QTL as one for which the distributions of alternative homozygotes for a particular QTL show little overlap, so that the probability of misclassification of phenotype is <5%, equivalent to 3.28 environmental standard deviations. For this study, we have represented QTL effects in terms of both the mean difference between parents and relative to the ESD, as we are most interested in whether substitution of alternative QTL alleles generates visible differences in phenotype relative to the two parents.

In this intraspecific study, we detected several sizeable QTL, the largest of which is on LG8. This QTL alone was responsible for divergence in floral, vegetative, and life-history traits. Although this and a few other fairly large QTL do not change the phenotype >3.28 ESDs, we argue that these QTL are major, particularly because they affect multiple traits. This QTL has a very large LOD score and a sharp peak, indicating that the QTL interval is fairly small, and it is very tightly linked to a single codominant marker, CYCB. A 2-LOD support interval around this QTL spans <10 cM (LG8: from 68 to 77 cM). Fine-scale mapping of this interval, followed by positional cloning may enable us to uncover the gene or genes responsible for divergence of multiple traits at this locus.

Most traits had at least one QTL that explained >10% of the species difference or that changed the phenotype >1 ESD, and the remaining QTL were of small effect. Overall, this is consistent with evolutionary predictions (Orr 1998) and with a few other empirical examples (Juenger et al. 2000; Perez-Perez et al. 2002) of the distribution of QTL effects, where the evolutionary shift in divergent characters between populations of Mimulus is likely to involve a major genetic change with most of the remaining divergence due to many minor allelic changes.

Pleiotropic QTL:

The floral, vegetative, and life-history traits measured in this study are governed largely by pleiotropic QTL. We define a pleiotropic QTL as a genomic region that affects multiple traits. This region could contain multiple tightly linked trait-specific genes or single genes that have multiple substitutions affecting different traits. Of the 28 total QTL we detected in this study, all but 3 affected multiple traits, pointing to a pleiotropic basis for the genetic associations that we observed (Tables 35). Fine-scale mapping with additional markers and larger mapping populations are needed to distinguish truly pleiotropic loci from tightly linked loci. Other studies have consistently found evidence suggesting that individual QTL have pleiotropic effects on multiple floral, vegetative, or life-history characters (Mitchell-Olds 1996; Juenger et al. 2000; Ungerer et al. 2002; Cui et al. 2004; Westerbergh and Doebley 2004), although all of these studies rely on QTL mapping on individual traits, rather than using our approach of joint mapping. The joint-mapping approach offers the advantage of allowing us to directly test whether different traits are affected by a particular QTL at that position and provides greater power to detect pleiotropy (Jiang and Zeng 1995). For this reason, previous studies may have underestimated the degree of pleiotropy.

For example, using joint QTL mapping, we identified 16 QTL underlying divergence in one or more of the six floral traits. Nearly all (14) of these QTL had significant effects on more than one floral trait for a total of 52 significant QTL–trait effects. If we had analyzed each floral trait in separate single-trait CIM analyses instead of in a joint-trait analysis, we would have detected 12 total floral QTL. Reliance on single-trait analysis has several limitations, on the basis of our results. First, it would have led us to overestimate the total number of QTL, because most of the 12 single-trait QTL would have mapped to the same genomic regions. Second, it has substantially less power to detect pleiotropic QTL than do joint mapping analyses (in our case a four- to fivefold difference in total QTL–trait effects); so we would have also grossly underestimated the total number of QTL–trait effects. For highly correlated traits, joint trait–QTL analysis provides a more comprehensive view of the genetic architecture underlying multivariate phenotypic divergence.

To better understand the role that pleiotropic QTL can have on our view of trait divergence, we examined the joint effect of the major QTL on LG8. In the floral traits analysis, this QTL had a large effect on both corolla tube length and stamen length. The direction of change at this QTL was almost perfectly correlated between traits (Figure 3A), demonstrating that substitution of this QTL alone into one parent can shift the phenotype roughly one-quarter of the way toward the alternate parent for both of these floral traits. Evolutionary divergence for either of these two floral traits is highly constrained. One might also expect overall “size” QTL to affect both floral and vegetative traits. We therefore examined the joint effect of this QTL on corolla tube length and stem thickness. These two traits are also positively correlated, although not as strongly as the two floral traits, and the QTL had a major effect on both traits (Figure 3B). On the basis of the effect of this QTL, flower size and plant size are likely to evolve jointly and in the same direction in Mimulus, which fits with a common observation that larger plants tend to produce larger flowers. Not all of the detected QTL affected traits in the same direction. We plotted the antagonistic effect of this major QTL on corolla tube length and leaf width. Increases in both of these traits are likely to be adaptive in both environments (Hall 2005), although the evolutionary response to selection operating on either trait will necessarily be constrained.

The large number of pleiotropic QTL detected in this study sheds light on our understanding of the genetic basis for multivariate divergence. If we were to examine each of these traits separately, we would misestimate the total number of QTL. Furthermore, any one QTL with modest individual effects on multiple traits can actually have a fairly large effect on the overall phenotype. Therefore a few pleiotropic QTL can play an important role in phenotypic divergence between populations or species.

The evolution of life-history strategy:

The divergence between the annual and perennial forms studied here is complex, consisting of differences in multiple floral, vegetative, and life-history characters. The life-history traits we measured were controlled by few QTL, suggesting that the evolution of differences in timing of flowering and allocation of floral and vegetative meristems would require only one or two genetic changes. For both life-history characters we measured, we showed that the genetic control of these individual traits is not independent of other morphological traits, which can have important implications for the potential for evolutionary divergence.

In QTL mapping experiments of other plant species, life-history variation is governed mainly by multiple genetic loci. In Arabidopsis thaliana, timing to flowering is controlled by 5–12 QTL (Mitchell-Olds 1996; Alonso-Blanco et al. 1998; Ungerer et al. 2002), and the particular QTL detected can differ when plants are grown in different environments (Weinig et al. 2002). In crop plants, traits differentiating annual and perennial forms are mostly polygenic (Paterson et al. 1995; Hu et al. 2003; Cui et al. 2004; Westerbergh and Doebley 2004). Without further genetic dissection of genomic regions in Mimulus and comparisons in different environments, it is difficult to determine if our results are inconsistent with polygenic inheritence of life-history traits.

In this study, we have begun to understand the genetic basis of phenotypic traits associated with life-history divergence. Future studies aimed at understanding how alleles at each QTL affect fitness in the wild will be particularly informative. We have developed recombinant inbred lines between these two divergent populations that have been placed into each of the native environments. Understanding the role of QTL genotype × environmental interactions and the particular QTL affecting fitness in these two diverse sites will further our understanding of the genetic basis of adaptation in the wild.


The authors thank M. Rausher, W. Morris, P. Manos, R. Vilgalys, A. Case, A. Sweigart, A. Cooley, Y.-W. Lee, D. Lowry, S. McDaniel, and J. Kelly for advice on earlier drafts of this manuscript. We also thank L. Fishman for advice on QTL mapping and for providing updated M. guttatus x M. nasutus map data. We give abundant thanks to E. Gilliam and K. Sullivan for assistance with quantitative measurements and to A. Bissell for graphical assistance. This material is based upon work supported by the National Science Foundation under grant nos. 9727578, 0075704, 0328636, and 010577; by Sigma Xi Grants-in-Aid-of-Research; and by the National Institutes of Health grant no. GM045344.


  • Communicating editor: S. W. Schaeffer

  • Received September 17, 2005.
  • Accepted November 22, 2005.


View Abstract