The formation of hybrid zones between nascent species is a widespread phenomenon. The evolutionary consequences of hybridization are influenced by numerous factors, including the action of natural selection on quantitative trait variation. Here we examine how the genetic basis of floral traits of two species of Louisiana Irises affects the extent of quantitative trait variation in their hybrids. Quantitative trait locus (QTL) mapping was used to assess the size (magnitude) of phenotypic effects of individual QTL, the degree to which QTL for different floral traits are colocalized, and the occurrence of mixed QTL effects. These aspects of quantitative genetic variation would be expected to influence (1) the number of genetic steps (in terms of QTL substitutions) separating the parental species phenotypes; (2) trait correlations; and (3) the potential for transgressive segregation in hybrid populations. Results indicate that some Louisiana Iris floral trait QTL have large effects and QTL for different traits tend to colocalize. Transgressive variation was observed for six of nine traits, despite the fact that mixed QTL effects influence few traits. Overall, our QTL results imply that the genetic basis of floral morphology and color traits might facilitate the maintenance of phenotypic divergence between Iris fulva and Iris brevicaulis, although a great deal of phenotypic variation was observed among hybrids.

The formation of hybrid populations is a widespread phenomenon that can be brought about by habitat disturbance (Anderson 1948), range expansion (Lewontin and Birch 1966; Williams and Arnold 2001), or invasion by exotic species (Abbott 1992; Anttila et al. 2000) as well as other ecological circumstances. Genetic mixing provided by hybridization can affect quantitative trait variation in hybridizing lineages, and hybridization of different crop lineages and related wild species has been widely used for crop improvement (examples in rice reviewed in McCouch 2004). Likewise, plant evolutionary biologists have long been interested in how hybridization may affect quantitative trait variation and the evolution of plant species (Anderson 1949; Ellstrand et al. 1996; Rieseberg 1997; Rieseberg et al. 2003b; reviewed in Arnold 1997, 2006). Numerous ideas have been developed concerning the possible evolutionary consequences of hybridization. For example, hybridization may: (1) produce novel trait combinations, producing lineages suited to novel combinations of ecological conditions (Anderson 1948; Howarth and Baum 2005); (2) allow the transference of an adaptive trait from one species to another (e.g., adaptive introgression; Anderson 1949); (3) lead to the formation of stable populations of hybrid genotypes phenotypically intermediate to the two parental species (e.g., Wu and Campbell 2006); (4) produce quantitative trait variation among hybrid genotypes that is extreme relative to the parental species (e.g., transgressive variation), facilitating the evolution of hybrid species inhabiting novel ecological habitats (e.g., Rieseberg et al. 2003b); or (5) lead to the loss of quantitative trait differentiation through genetic assimilation (e.g., Johnsgard 1967; Mank et al. 2004). Hybridization is also the first step in the formation of allopolyploid species (Briggs and Walters 2001).

The way that hybridization affects quantitative trait variation, introgression, and subsequently quantitative trait evolution will depend in part on the genetic basis of the traits involved. By genetics we mean such aspects as the number of loci influencing trait variation, the size or magnitude of individual locus' phenotypic effects, and the extent of linkage between loci affecting different traits or contributing to species incompatibilities. Quantitative trait locus (QTL) mapping provides a method of describing the number, phenotypic effects, and genomic locations of genetic factors contributing to quantitative trait variation (Lander and Botstein 1989; Tanksley 1993; Zeng 1994; Mauricio 2001). QTL mapping has been widely used by evolutionary biologists during the last decade to study the genetic basis of phenotypic divergence during the process of speciation (e.g., Bradshaw et al. 1995; Schemske and Bradshaw 1999; Hawthorne and Via 2001; Fishman et al. 2002; Rieseberg and Wendel 2004; Lexer et al. 2005; Martin et al. 2007). These studies of speciation have focused on three aspects of the genetic basis of quantitative traits: the size or magnitude of individual QTL effects, the colocalization of QTL for different traits, and the proportion of mixed QTL effects (see below). Each of these aspects, as well as the number of loci affecting a trait, is theoretically expected to affect the evolutionary dynamics of phenotypic divergence and speciation, such as the tempo of divergence, the feasibility of sympatric speciation, and the likelihood of correlated evolution among different traits (Dobzhansky 1937; Smith 1983; Barton and Charlesworth 1984; Gottlieb 1984; Coyne and Orr 1998; Kondrashov and Kondrashov 1999; Hawthorne and Via 2001; Rieseberg et al. 2003a). These three aspects of quantitative genetics would also affect the extent and pattern of quantitative trait variation in hybrid zones, and thus their evolutionary potential. This article reports the results of a QTL mapping study designed to examine the genetic basis of floral trait variation of hybrid populations formed by the intermating of two species of Louisiana Irises.

The first of these aspects—the size or magnitude of individual QTL effects—can affect the tempo of evolutionary divergence and speciation: if QTL alleles have a large phenotypic affect (i.e., large relative to the phenotypic difference between species), then perhaps few genetic changes were required for the evolution of the traits characteristic of the species (Gottlieb 1984; Bradshaw et al. 1995; Coyne and Orr 1998). QTL studies have shown that species differences can be due to alleles with large phenotypic effects (Bradshaw et al. 1995; Schemske and Bradshaw 1999). In the context of a hybrid zone population, this means that the introgression of a single allele can have a large ecological impact (e.g., Bradshaw and Schemske 2003). Large-effect QTL alleles also mean that divergent natural selection could more easily reassemble parental species phenotypes from a hybrid population than if numerous genetic factors of small phenotypic effect are involved (Fishman et al. 2002; Lexer et al. 2005).

The second aspect of quantitative genetics, the colocalization of QTL for different traits, is expected to impact the evolution of species differences involving multiple traits (Hawthorne and Via 2001; Rieseberg et al. 2003b; Via and Hawthorne 2005). Adaptive species differences often comprise a complex of functionally related traits, such as the combination of flower morphology, orientation, and color that adapts a plant to service by a particular pollinator (e.g., Fulton and Hodges 1999; Hodges et al. 2002). If such functionally related traits are genetically correlated due to pleiotropy or tight linkage, this may facilitate a population's response to selection acting jointly on these traits (Lande 1979; Lande and Arnold 1983). In hybrid populations, genetic correlations should cause trait correlations in hybrids to resemble those of the parental species, facilitating the maintenance of species divergence (Hawthorne and Via 2001; Via and Hawthorne 2005). If instead trait correlations in hybridizing species are not due to genetic correlations, but rather due to linkage disequilibrium, then recombination will be more likely to create novel trait combinations in hybrids (Rieseberg et al. 2003b; Via and Hawthorne 2005).

The third aspect of quantitative trait variation is the number of mixed genetic affects harbored by the hybridizing species (Orr 1998; Rieseberg et al. 2002; Griswold and Whitlock 2003; Rieseberg et al. 2003a). For example, imagine a QTL mapping experiment involving a tall and a short species of plant that uncovers seven QTL segregating in the tall line, where five of these QTL increase plant height and two decrease it. This is a case of mixed QTL effects, because two out of seven of the QTL had effects that were “opposite” or “negative” with respect to expectation, based on the phenotype of the tall species. A lack of mixed effects would be a case in which of the seven QTL, all increase plant height. The proportion of mixed QTL effects underlying a trait is of interest because it is theoretically expected to reflect the relative strength of directional selection versus drift that acted during the evolution of that trait (Orr 1998; Rieseberg et al. 2002; but see Anderson and Slatkin 2003). The occurrence of mixed QTL effects is also important in the context of hybrid populations, where the mix of QTL effects from the hybridizing species can be recombined to produce transgressive variation in hybrids. Transgressive variation is defined as variation in segregating hybrid populations that includes extreme phenotypes that exceed or subseed those of the parents (Devicente and Tanksley 1993; Cosse et al. 1995; Rieseberg and Linder 1999; Rieseberg et al. 2003a). Hybrid populations produced from the intermating of species that harbor QTL of mixed effects have the potential for greater phenotypic diversity than hybrid populations formed from species without mixed QTL effects (Lexer et al. 2005).

This article describes a QTL mapping study in which the above three aspects of quantitative trait variation were examined in two species of Louisiana Irises. These two species differ in numerous floral traits comprising alternative pollination syndromes, yet form natural hybrid zones in southern Louisiana. Iris fulva and I. brevicaulis are strikingly different in floral form and are primarily pollinated by different animals (Viosca 1935; Wesselingh and Arnold 2000). Iris brevicaulis is primarily bumblebee-pollinated (Wesselingh and Arnold 2000) and possesses a suite of traits associated with a bee-pollination syndrome: blue flowers marked with prominent white and yellow nectar guides, stiff upright sepals, and strongly scented flowers. Iris fulva is primarily hummingbird-pollinated (Emms and Arnold 2000; Wesselingh and Arnold 2000) and possesses many traits characteristic of a hummingbird-pollination syndrome: red flowers with protruding anthers and highly reflexed sepals. Iris fulva and I. brevicaulis are broadly sympatric throughout the Mississippi River valley of central North America, but are found in different habitats (Viosca 1935; Cruzan and Arnold 1993; Johnston et al. 2001). Iris fulva is found in intermittently flooded, forested wetlands throughout this range, whereas I. brevicaulis is normally found in drier, riparian-associated hardwood forests. These species also differ in flowering time: I. fulva populations begin flowering ca. one month earlier than I. brevicaulis populations in southern Louisiana (Cruzan and Arnold 1993). Despite the differences in habitat, floral morphology, and phenology, these species form extensive hybrid zones in southern Louisiana, often as a consequence of habitat disturbance (Viosca 1935; Arnold et al. 1992; Cruzan and Arnold 1993; Johnston et al. 2001). Hybrids observed in these populations both phenotypically and genetically resemble advanced generation backcross hybrids (Arnold et al. 1992; Cruzan and Arnold 1993; Johnston et al. 2001).

In this study, QTL mapping was used to investigate the genetic basis of the divergent pollination syndromes of I. fulva, I. brevicaulis, and interspecific hybrids. Variation in nine morphological and coloration traits comprising these pollination syndromes was characterized in reciprocal backcross (BC1) hybrids in a greenhouse environment. QTL mapping was used to assess the size of QTL effects underlying these traits, allowing us to determine the number of QTL substitutions separating the parental species' phenotypes, and assess the phenotypic effects of QTL introgression. Results were also used to determine whether QTL underlying different traits were found to colocalize, which would imply a genetic basis for the maintenance of parental-species-like trait correlations in hybrid populations. In addition, the occurrence of mixed QTL effects was examined to assess the likelihood of transgressive trait variation in hybrid populations.

Materials and Methods


Construction of reciprocal I. fulva×I. brevicaulis BC1 mapping populations are described elsewhere in detail (Bouck et al. 2005). Briefly, two wild-collected individuals (I. fulva genotype If174 and I. brevicaulis genotype Ib72) were used to make reciprocal interspecific BC1 mapping populations. The same individuals were used as both the F1 parents and as the recurrent parents for backcrossing. This was done to minimize the within-species quantitative genetic variation segregating in the experiment. The individuals used were collected from natural populations in southern Louisiana that had apparently not experienced hybridization with the other species, as determined by population genetic analysis or by population observations. The I. fulva individual, If174, was collected from a population in Terrebonne Parish, and the I. brevicaulis individual, Ib72, was collected from a population in St. Martin Parish. Flowers of Ib72 were crossed with pollen from If174 to make F1 hybrids in 1997. Two F1 hybrids, designated F1(2) and F1(3), were used to make backcross hybrids during the winter of 1999. Pollen from F1(3) was crossed onto flowers of several ramets (clones) of Ib72 to make I. brevicaulis BC1 hybrids (IbBC1), and pollen from F1(2) was crossed onto flowers of several ramets of If174 to make I. fulva BC1 hybrids (IfBC1). All flowers were emasculated prior to opening, and pollen was applied two days later after the stigmatic surface became receptive. All plants were housed in the Department of Plant Biology greenhouses at the University of Georgia. Plants typically began flowering in January of each year and ceased flowering by early May.


Morphological and color traits distinguishing I. fulva and I. brevicaulis were identified as components of the I. fulva hummingbird and I. brevicaulis bumblebee pollination syndromes (Fig. 1, Table 1). These traits were measured in 2001, 2002 and 2003 on multiple ramets (clones) of the Ib72, If174, F1(2), and F1(3) parents as described below. The mean, variance, and standard error of each trait were calculated for each parental genotype and used to calculate the difference in species means (DSM or species difference), the environmental variance (VE) and environmental standard deviation (ESD). The ESD reflects the magnitude to which a trait varies solely due to environmental influence, and pools variation both between years and between ramets within years. The ESD is the square root of VE, which is calculated as a weighted average of the parental phenotypic variances (e.g., the variance among clones; Lynch and Walsh 1998)

Figure 1.

Flowers of I. brevicaulis (A and B) and I. fulva (C and D). Panel E shows the sepals of I. fulva (left) and I. brevicaulis (right) dissected at the base of the calyx. See text for explanation of traits.

Table 1.  Components of the I. fulva and I. brevicaulis pollination syndromes. Parental genotype mean trait values and standard errors are indicated on the top row of each cell, sample sizes (in parentheses) and variances are on the bottom row. The DSM is the difference in the species' mean trait values, the ESD is the environmental standard deviation of a trait. See text for units and description of traits.
Component traitabbreviation Ib72If174F1(2)F1(3)DSMESDDSM/ESD
Flower stalk height relative to leavesSTALK0.51±0.021.41+0.070.76+ 0.030.90+0.03−0.900.13−6.92
(18) 0.01(7) 0.04(15) 0.01(18) 0.02 
Anther extensionANTHEX−6.35±0.192.58+0.27−3.05+0.22−2.12+0.25−8.930.96−9.30
(29) 1.08(11) 0.85(17) 0.79(24) 1.03 
Stylar branch length SBL31.18±0.3819.86+0.1326.87+0.5226.42+0.20  11.321.53   7.40
(9) 1.28(9) 0.15(21) 5.72(15) 0.59 
Nectar guide areaNGA58.22±2.58036.82+3.6820.27+1.96  58.2216.98   3.43
(30) 220.00 (17) 229.95(24) 92.63 
Sepal total lengthSEPTL62.49±0.6954.00+1.4469.77+0.9059.50+0.72   8.493.03   2.80
(9) 4.33(9) 18.69(21) 9.76(15) 6.21 
Sepal shapeSPAT0.79±0.000.38+0.020.61+0.010.59+0.02   0.410.05   8.20
(9) 0.00(9) 0.00(21) 0.00(15) 0.00 
Sepal blade chromaCHROMA6.9844±0.29574.5796+0.78506.97+0.25495.79+0.1566   2.2940.76   3.02
(10) 0.3496(6) 1.8482(8) 0.5199(5) 0.1225 
Sepal blade hueHUE−0.8028±0.0578−0.6559+0.1199−0.9587+0.2549−0.8509+0.0293−0.14690.1257−1.1683
(10) 0.0133(6) 0.0431(8) 0.0149(5) 0.0043 
Sepal blade brightnessBRIGHT0.2910±0.01760.2119+0.01680.2049+0.00370.1574+0.00700.07910.0216  3.66
(10) 0.0012(6) 0.0008(8) 0.0001(5) 0.0002 


Backcross genotypes did not flower every year, and some genotypes did not flower at all during the course of this study. As a result, phenotypic data were collected in 2001, 2002, and 2003. Data for four of the traits (stylar branch length, sepal stalk length, sepal blade length, and sepal total length) were collected solely in 2003. Phenotypes were measured on 179 IbBC1 and 75 IfBC1 genotypes (Table 2).

Table 2.  Sample size (N) and Spearman's P correlation coefficients for tests of pairwise trait correlations. Correlations are between traits within a backcross population. Values for the IfBC1 hybrids are below and to the left of the diagonal, those for the IbBC1 hybrids are above and to the right. Correlations significant after Bonferroni adjustment are shown in bold.
 IbBC1 hybrids
N 198 176 189 132 132 179 179 179 179
IfBC1 hybrids 
  STALK87    0.1385−0.1817−0.0229   0.0875   0.1670   0.2740   0.1223−0.2339
  SBL71−0.1017 0.3092−0.0084   0.2702   0.0682   0.0354−0.0935−0.1284
  ANTHEX81−0.08940.5207 −0.2325   0.1006−0.1023−0.2600−0.0741   0.2557
  SPAT56−0.0797−0.2465   0.2709 0.45270.4401   0.2484   0.0839−0.2374
  SEPTL56−0.0832   0.3806   0.0174−0.2799    0.0335−0.4850−0.0770   0.1873
  NGA80   0.1792−0.0233   0.0666−0.0551   0.113    0.4008   0.0548−0.2928
  BRIGHT75   0.1821   0.0112−0.0596   0.0484   0.1689   0.1374    0.24390.5412
  CHROM75   0.0179   0.2277−0.1019−0.2452   0.3135   0.3216   0.2897 0.5636
  HUE75−0.2733−0.1170   0.1569−0.0623−0.1847−0.1391   0.0107−0.2788 

Morphological and coloration traits were measured on the first apical flower of each genotype ca. 8 h after opening (Table 1). Flower stalk height (STALK; measured at the base of the calyx of the apical flower: the base of the calyx is the base of the sepals and petals) and plant height (height of the tallest rhizomatous leaf) were measured to the nearest centimeter. For the remaining traits, all three floral units of each flower were measured (Iris flowers are tripartite), and these values were averaged for each genotype.

The following morphological traits were measured to the nearest 0.01 mm using digital calipers: (1) anther extension (ANTHEX) was the distance the tip of the anther extended past the end of the stylar branch, measured on intact flowers; (2) stylar branch length (SBL) was measured from the base of the calyx to the distal end of the stigma, also measured on intact flowers; (3) sepal total length (SEPTL); (4) sepal stalk length; and (5) sepal blade length were measured after dissection from the base of the calyx. Sepal total length was the length of the entire sepal. Sepal stalk length was the length of the sepal from the base of the calyx to the inflection point in the curve in which the sepal stalk flares out into the blade (see Fig. 1). Sepal blade length is the distance from this inflection point to the distal end of the sepal. The ratio of the sepal stalk length to the sepal blade length provides a measure of the degree to which sepal shapes (SPAT) vary from spatulate (I. brevicaulis-like) to pendate (I. fulva-like) (Fig. 1). Sepal shape corresponds to the degree of reflexivity of the sepals of intact flowers (Fig. 1).

Coloration traits were measured as follows: The length and width of the visible area of contrasting coloration (the nectar guide) on the sepals of intact flowers was measured using digital calipers. As nectar guides on Iris flowers are roughly triangular, the nectar guide area (NGA) was calculated as one half the length times the width of this area. Sepal blade color was quantified by calculating a standardized reflectance spectrum with a fiber optics reflectance spectrometer (Unispec Spectral Analysis system, PP Systems, Amesbury MA). This uses a built-in halogen light source to illuminate the adaxial surface of sepal tissue enclosed in a standard cuvette clipped onto the sepal near the distal end. The spectrometer records the average of 10 measurements of reflectance at light wavelengths from 350 to 1100 nm. The average reflectance spectra of each sepal was standardized against the reflectance spectra of a standard, and reflectance spectra from all three sepals were averaged to produce a genotype mean. Reflectance data were used to calculate the proportion of total brightness occurring in four equal portions of the visible light spectrum: 400–475 nm (blue), 475–550 nm (green), 550–625 nm (yellow), and 625–700 nm (red), and these values were used to calculate the chroma, hue and brightness of each genotype (after Endler 1990; Hodges et al. 2002). Chroma (or saturation) was calculated as inline image. Hue (or color) was calculated as arcsine[inline image]. Brightness was calculated as the proportion of total incident light reflected from 400 to 700 nm.


Transgressive hybrid phenotypes were detected by comparing the number of hybrid progeny with trait values subseeding or exceeding parental mean trait values by 2 standard deviations to the number expected by chance (after Rieseberg et al. 2003b).


Nonparametric measures of correlation (Spearman's ρ) were estimated for all pairwise trait combinations separately for each BC1 mapping population using JMP statistical software (the SAS Institute 2005). Statistical significance was evaluated after Bonferroni correction for multiple comparisons (Sokal and Rohlf 1994).


Molecular marker data collection and genetic map construction are described elsewhere in detail (Bouck et al. 2005). The I. fulva linkage map consists of 142 framework markers spanning 1439 Kosambi cM on 22 linkage groups, with a coverage estimate of 76% of the genome within 10 cM of a framework marker (Bouck et al. 2005). The I. brevicaulis linkage map consists of 106 markers spanning 1109 Kosambi cM on 22 linkage groups, resulting in a coverage estimate of 72–74% of the genome within 10 cM of a framework marker (Bouck et al. 2005). Note that IRRE transposon display markers are dominant and that the homology of linkage groups between the I. fulva and I. brevicaulis maps has not yet been established.


Genome scans for quantitative trait loci were performed with the composite interval mapping (CIM) method of Zeng (1993; 1994), implemented by the software program QTL Cartographer version 1.17d (Basten et al. 1994, 2004). CIM combines interval mapping with multiple regression to test for the presence of a QTL conditional on the effects of other QTL (cofactors) also affecting the trait. Each trait and each mapping population was analyzed separately. Forward and backward stepwise regression with a critical value of 0.05 was used to identify up to five cofactors subsequently used in the CIM analysis. CIM was used to calculate a test statistic at 2-cM intervals, excluding the effects of cofactors within a 10 cM sliding window. For each trait, a genome-wide significance threshold corresponding to a Type 1 error rate of 0.05 was determined by performing 1000 permutations (Churchill and Doerge 1994; Doerge and Churchill 1996). QTL locations were reported as the map location at which the LR test statistic was greatest, and 2-LOD confidence intervals for QTL location were calculated from the CIM results. QTL effects were estimated by the CIM procedure, and reported in terms of the percent variation explained (PVE) in the mapping population, scaled by ESDs, and as a proportion of the DSM.


A permutation approach was used to determine the significance of observed QTL colocalization. QTL locations were assumed to be random and independent. The total map length was divided into intervals roughly equal in length to the average 2-LOD confidence intervals for mapped QTL (40 cM). The total number of QTL detected were then randomly distributed among these intervals, and the greatest number of QTL located in a single interval was recorded. This process was repeated 1000 times to empirically determine the 95% confidence value for the average number of QTL expected to occur in a single map interval by chance.



The I. fulva and I. brevicaulis parents were differentiated for all floral morphology and coloration traits measured (Table 1; Fig. 1). For each trait, the mean difference between the I. brevicaulis and I. fulva parent (DSM) was scaled relative to the observed ESD of the trait. The ratio of the DSM relative to the ESD provides an assessment of the degree to which I. brevicaulis and I. fulva parents differ beyond the range of variation typically induced by environmental noise.

The species differences ranged from 1.78 to 9.30 ESDs, suggesting that there is significant genetic differentiation between I. brevicaulis and I. fulva at all of these traits (Doerge et al. 1997). Floral parts of I. brevicaulis tended to be larger overall than those of I. fulva flowers, with the exception of the sepal blade length (Table 1). Parental species flowers also differed in coloration: the average chroma of I. brevicaulis flowers was 3.02 ESDs greater than that of I. fulva, and I. brevicaulis flowers were on average 3.66 ESDs brighter than those of I. fulva. Parental species flowers also differed in hue by 1.17 ESDs.

The F1 hybrids exhibited trait values that were generally intermediate between the two parents, with some exceptions (Table 1). The F1(2) parent but not the F1(3) parent exceeded the larger I. fulva parent in sepal blade length. Both F1 hybrids subceeded both parents in brightness.


The range of phenotypic variation in both BC1 populations was considerable, and transgressive hybrid phenotypes were observed in at least one BC1 population for SBL, NGA, SEPTL, SPAT, HUE and BRIGHT (Fig. 2). Few trait combinations were significantly correlated after Bonferroni correction, but those that were generally included measures of shape or relative size and their component parts (e.g., SPAT and SEPTL, ANTHEX and SBL: Table 2), or other traits that seem likely to be developmentally correlated (e.g., NGA and SPAT). Several coloration traits were significantly correlated in IbBC1 but not IfBC1 hybrids (e.g., BRIGHT and NGA, HUE and BRIGHT, and HUE and CHROMA).

Figure 2.

Phenotypic variation in Iris hybrids. Histograms show the distribution of trait variation in IbBC1 hybrids (gray graphs) and IfBC1 hybrids (white graphs). Whisker plots below each histogram span 2 standard deviations above and below the recurrent parental mean. Traits for which there is transgressive variation are marked with a “T.” See Table 2 for sample sizes, and materials and methods for explanation of traits, test for transgressive variation, and units of measurement.


A total of 10 QTL were found for the nine traits analyzed (Table 3A). QTL effect sizes ranged from 8% to 69% of the species difference. These QTL were located on six of the 22 chromosomes in the linkage map. QTL affecting BRIGHT, NGA, HUE, SPAT and ANTHEX were clustered in a roughly 40-cM interval on linkage group 4 (Fig. 3A, Table 3A). Based on permutation tests of this result, the colocalization of four of 10 QTL within a single 40 cM map interval would be expected to occur less than 99% of the time by chance alone. It is possible that recombinational distances in this genomic region might be different in intraspecific crosses, or even in different interspecific crosses. The majority of traits did not have QTL with mixed phenotypic effects. A single trait, STALK, had QTL with phenotypic effects that were “opposite”: these QTL act to make flower stalks shorter, even though I. fulva is the parent with markedly taller flower stalks (Table 1).

Table 3.  QTL underlying components of the I. fulva and I. brevicaulis pollination syndromes. See text for description of traits. Sample sizes (N) for each trait are listed. QTL locations by linkage group (LG), map position estimates (in Kosambi cM) and associated 2 LOD confidence intervals (CI) are given. QTL effects (1a) were estimated by CIM analysis. QTL effect sizes are also given as absolute values scaled in three ways: as a proportion of the difference in species means (DSM), in units of environmental standard deviation (ESD), and by the percent variation explained (PVE: expressed here as a proportion). QTL effects in the expected phenotypic direction are indicated in bold type; QTL effects opposite the expected direction are shown in italics. For traits with more than 1 QTL, the average QTL effect for each trait is given, as well as the overall average QTL effect for all traits.
A. Iris fulva QTL
TraitNQTL locationSignificanceDirectionSize
LGPosition2 LOD CI0.05 sig. levelTest Stat.est. effect (1a)by DSMby ESDPVE
STALK198 663.2952–7412.691817.5622−0.07210.08010.55460.0709
ANTHEX189 440.5428–5713.261214.0386   1.11960.12541.16630.0744
SPAT132 434.54 8–4412.906724.84340.07800.19021.56000.1620
21 7.26 0–13 16.33320.05730.13981.14600.0865
 trait ave.:0.16501.35300.1243
NGA179 416.01 2–4213.569432.520524.31400.41761.43190.2586
1410.41 0–43 14.229312.78350.21960.75290.0734
 trait ave.:0.31861.09240.1660
BRIGHT179 418.01 4–4413.395556.64700.05490.69412.54170.3697
 230.6610–55 16.27110.02760.34891.27780.0927
 trait ave.:0.52151.90970.2312
HUE179 426.01 0–4411.230011.4510   0.08940.60860.71120.0917
 397.884–113 12.1800   0.06670.45410.53060.0650
 trait ave.:0.53130.62090.0784
 ave. over all traits: 0.32781.16730.1345
B. Iris fulva QTL
TraitNQTL locationSignificanceDirectionSize
LGPosition2 LOD CI0.05 sig. levelTest Stat.est. effect (1a)by DSMby ESDPVE
STALK87NO QTL 14.1371 
SBL71NO QTL 12.8846 
ANTHEX811017.24  2–1813.034815.28371.24790.13971.29990.1587
SPAT56NO QTL 12.9836 
SEPTL56 855.79 49–6913.425221.3845   7.19420.84742.37430.2554
NGA80NO QTL 12.7220 
BRIGHT75 2 6.01  6–2113.797935.9976−0.05510.69662.55090.3309
CHROM75 4 0.01  0–1012.590212.9434   0.45480.19830.59840.1244
HUE75 1184.74162–20013.129415.59860.28751.95712.28720.2202
 ave. over all traits:0.76781.82220.2179
Figure 3.

QTL locations for I. fulva (A) and I. brevicaulis (B) floral traits. Likelihood ratio (LR) plots are given for all nine traits across both genetic maps. Map distances on the x-axis are given in Kosambi cM: LR is on the y-axis. Asterisks indicate LR peaks representing significant QTL. See Table 3 for individual trait significance thresholds. Iris brevicaulis linkage group 22 is not shown (it is 0 cM in length).


Only five I. brevicaulis QTL were detected, possibly due to the much smaller sample sizes used for these analyses (Table 3B). In fact it is likely that due to the small sample sizes, QTL with smaller phenotypic effects could not be detected. QTL effects ranged in magnitude from 14% to 196% of the species difference. These QTL were all on separate linkage groups: no colocalization of QTL was observed (Fig. 3B, Table 3B). Only one trait, brightness (BRIGHT), was influenced by a QTL with “opposite” phenotypic effect: the single QTL for brightness on linkage group 2 decreases sepal brightness, even though I. brevicaulis is the parent with brighter sepals (Table 1).



This study was designed to assess how the genetic basis of traits might govern quantitative trait variation in hybrid populations, such as those formed in nature between I. brevicaulis and I. fulva (Viosca 1935; Arnold et al. 1992; Johnston et al. 2001). Hybrids of I. fulva and I. brevicaulis demonstrate considerable variation in floral traits related to pollination syndrome (Fig. 2), and the QTL underlying this variation are typically of large phenotypic effect, with values equal to roughly 30% of the phenotypic species difference being typical (Table 3). This implies that relatively few genetic steps, in terms of QTL substitutions, separate these two species' phenotypes. However, our estimates of QTL effects must be interpreted with caution due to small sample sizes, which are likely to lead to an overestimation of the effects of detected QTL (Tanksley 1993; Beavis 1994; Doerge et al. 1997). This (the Beavis effect) may in part explain some of the extremely large QTL effect estimates listed in Table 3.

The result of this and other studies reporting QTL of large effect underlying species differences support the conclusions of Gottlieb (1984), who argued that genes with large phenotypic effects underlie major quantitative morphological differences in plants. Likewise, Orr and Coyne (1992) later came to agree with this view, stating that genes of large effect are likely to be important in adaptive speciation. Indeed, numerous cases of QTL with large phenotypic effect have been found for yield and structural characteristics of domesticated crop species (Doebley and Stec 1993; examples in Tanksley 1993; Grandillo and Tanksley 1996; Frary et al. 2000; Liu et al. 2002; van der Knaap and Tanksley 2003). Studies of adaptive species difference in wild plants have also uncovered QTL of large effect. Bradshaw et al. (1995,1998) found QTL with large phenotypic effects segregating in two closely related species of Mimulus that have contrasting hummingbird and bumblebee pollination syndromes like the species of Iris studied here. In wild sunflowers, the QTL effects (2a) for 22 morphological traits differentiating Helianthus petiolaris and H. annuus were on average 79% of the species difference, with a range of 10% to over 100% of the species difference and a majority under 50% (Rieseberg et al. 2003b, see data in tables 1 and S1). Westerbergh and Doebley (2002) examined differences in tassel (male inflorescence) morphology distinguishing two wild species of Teosinte (Zea diploperennis and Z. mays ssp. parviglumis), and found that the average effect size of QTL was 28% of the DSM, with a range of 4% to 108% and a majority under 25%. In contrast to the above examples, two additional studies examining the genetic basis of floral differences between outcrossing and selfing species of Mimulus reported generally smaller QTL effect sizes: Fishman et al. (2002) found that the majority of floral trait QTL had effects smaller than 10% of the species difference, and Lin and Ritland (1997) reported typical QTL effects of 10–20% of the DSM.


A total of 10 I. fulva QTL and five I. brevicaulis QTL were associated with nine floral traits. Iris fulva QTL for different traits were located within the same genomic region, indicating some extent of genetic correlation between these traits. Traits that are genetically correlated due to tight linkage or pleiotropy of QTL will likely remain correlated in hybrid populations (Hawthorne and Via 2001; Rieseberg et al. 2003b; Via and Hawthorne 2005). In contrast, if trait correlations in the parental species are due to linkage disequilibrium between unlinked QTL, these trait combinations can be broken apart through recombination in a hybrid zone (Via and Hawthorne 2005). In IbBC1 hybrids, QTL (i.e., I. fulva QTL) for five floral traits (BRIGHT, NGA, HUE, SPAT, and ANTHEX) colocalized in a roughly 40-cM interval on linkage group 4 (Fig. 3A, Table 3A). The colocalization of these QTL implies that trait correlations in hybrid populations will likely resemble those of the parent species, and may remain through a few generations of recombination. In line with this prediction, four of these traits (BRIGHT, NGA, HUE, and SPAT) were significantly correlated in the IbBC1 mapping population (Table 2).

Colocalization of QTL was not observed for I. brevicaulis QTL, but the small number of QTL resolved in the IfBC1 backcross hybrids makes it difficult to assess whether QTL demonstrate colocalization. Furthermore, due to small sample sizes, the number of QTL resolved in this study is likely to be an underestimate of the actual number of genetic factors affecting variation in floral traits (Tanksley 1993; Doerge et al. 1997), thus contributing to lower power for detecting QTL colocalization.


Only two traits, STALK and BRIGHT, were affected by QTL with phenotypic effects opposite to the expected direction given the phenotypes of the I. brevicaulis and I. fulva parents (Table 3), suggesting that there is little potential for transgressive variation in hybrid populations of Louisiana Irises. This is because transgressive variation in hybrid populations has most often been shown to be due to the segregation of mixed QTL effects harbored by the parental species (Rieseberg et al. 2003a). Recombination of these QTL can produce hybrid genotypes that combine, for example, all the QTL from both parental species that increase a given trait, producing some hybrid individuals that are phenotypically extreme. Results from previous studies have supported a relationship between the proportion of negative, or mixed, effect QTL underlying a trait and transgressive variation in segregating hybrids (Davies 1971; Davies and Workman 1971; Devicente and Tanksley 1993; Rieseberg et al. 2003b; Rieseberg et al. 2003a).

In this study, the lone I. brevicaulis QTL of negative effect observed was for BRIGHT, and transgressive hybrid phenotypes were in fact observed for this trait (Fig. 2, Table 3B). However, four additional traits for which QTL of mixed effect were not observed also demonstrated transgressive variation in hybrids (SBL, SEPTL, SPAT, and HUE: Fig. 2). Another possible explanation for the transgressive variation observed in these hybrid populations is the segregation of quantitative genetic factors originating from the recurrent parents of each cross. The Ib72 and If174 genotypes used as both parents of the F1 hybrids and as recurrent backcross parents were wild-collected plants, with observed heterozygosities (He) of 0.59 and 0.31, respectively (Bouck et al. 2005). Even though an inbred crossing design was used, this means that considerable allelic variation was still harbored by the parental lines and segregating in the mapping populations. This inbred crossing design may have also contributed an additional source of transgressive variation: inbreeding has been shown to lead to the expression of rare recessive alleles, causing transgressive variation in hybrids (Rick and Smith 1953). Although these two possible explanations cannot adequately be assessed with the experimental design employed in this study, there are undoubtedly some additional sources of quantitative trait variation impacting the mapping populations studied here, because for all the traits, the total QTL effects explain far less than 50% of the total variation observed (see PVE column in Table 3).

The results of this study are subject to several caveats due to both the nature of QTL studies and our relatively small sample sizes. First, it is unknown at this point as to how generalizable our findings may be: one of the limitations of QTL studies in general is that the results may be specific to the individual genotypes used to generate the mapping populations. In addition, our sample sizes were limited by the long generation time and perennial life history of Iris, which means that genotypes may not flower on a consistent, yearly basis. It is true that other QTL studies of species differences in plants are based on larger samples sizes than our study. For example, Bradshaw et al.'s (1998) QTL mapping study of Mimulus was based on a mapping population of 465, Rieseberg et al.'s (2003b) study of Helianthus had a mapping population of 400, and Fishman et al.'s (2002) QTL analysis was based on a mapping population of 500. It is likely that with a larger sample of hybrids, we would identify additional QTL. And as was mentioned above, our relatively small sample size may have also led to inflated estimates of QTL effect sizes.

The results of this study of the genetic basis of floral trait variation in hybrids of I. fulva and I. brevicaulis suggest that natural hybrid zones should be characterized by parental-like phenotypes. QTL for five of the nine traits examined were colocalized on linkage group 4 in the IbBC1 hybrid population. This should lead to the joint genetic transmission of these traits in the early generations of hybrid zone formation, and trait combinations in hybrids should thus resemble those of the parent species. In addition, the fact that mixed QTL effects influence few traits predicts that there is little potential for transgressive variation in hybrid populations. However, counter to this expectation, transgressive variation was observed for six out of nine traits, indicating that considerable quantitative trait variation is produced via hybridization, even though the mechanism for this remains unclear. Studies are currently underway to assess the ecological impact of floral trait variation in hybrids in interactions with pollinators under natural field conditions (Martin et al., unpubl. data).

Associate Editor: Meagher 


We thank R. Mauricio for providing invaluable advice throughout the course of this project, and Z.-B. Zeng for generously taking the time to discuss with us the intricacies of QTL mapping. M. Boyd and A. Tull provided expert plant care. R. Peeler, R. Koopman, E. Morgan, and N. Schell assisted in floral measurements. This manuscript benefited from critical review by G. Baucom, M. Campbell, and members of the Vision lab at UNC Chapel Hill. A National Science Foundation Predoctoral Fellowship, the National Institutes of Health, and a University of Georgia Dissertation Completion Award supported AB. SRW was supported by the National Institutes of Health and the National Science Foundation. MLA was supported by National Science Foundation grants DEB-0074159 and DEB-0345123.