GENETIC ARCHITECTURE OF SEXUAL DIMORPHISM IN A SUBDIOECIOUS PLANT WITH A PROTO-SEX CHROMOSOME

Authors


Abstract

The rise of sexual dimorphism is thought to coincide with the evolution of sex chromosomes. Yet because sex chromosomes in many species are ancient, we lack empirical evidence of the earliest stages of this transition. We use QTL analysis to examine the genetic architecture of sexual dimorphism in subdioecious octoploid Fragaria virginiana. We demonstrate that the region housing the male-function locus controls the majority of quantitative variation in proportion fruit set, confirming the existence of a proto-sex chromosome, and houses major QTL for eight additional sexually dimorphic traits, consistent with theory and data from animals and plants with more advanced sex chromosomes. We also detected autosomal QTL, demonstrating contributions to phenotypic variation in sexually dimorphic traits outside the sex-determining region. Moreover, for proportion seed set we found significant epistatic interactions between autosomal QTL and the male-function locus, indicating sex-limited QTL. We identified linked QTL reflecting trade-offs between male and female traits expected from theory and positive integration of male traits. These findings indicate the potential for the evolution of greater sexual dimorphism. Involvement of linkage groups homeologous to the proto-sex chromosome in these correlations reflects the polyploid origin of F. virginiana and raises the possibility that chromosomes in this homeologous group were predisposed to become the sex chromosome.

Phenotypic differences between the sexes (sexual dimorphism) can be dramatic, but just how these arise in a common genome is a long-standing question. Sexual dimorphism (SD) is thought to arise from selection toward different trait optima in the sexes (Lande 1980). At the genetic level, one way SD can arise is through linkage between the genes with sex-differential fitness effects, that is, sexually antagonistic genes, and the sex-determining region. This mechanism is thought to prevail because the conditions for the spread of sexually antagonistic genes in a population are far less stringent when linked to the sex-determining region than when on autosomes (Fisher 1931; Charlesworth and Charlesworth 1980; Bull 1983; Rice 1984, 1992). A contrasting view, however, holds that autosomal genes are the key players in controlling SD through sex-limited or sex-biased expression, in part because a majority of traits are polygenic and hence sex-linkage ought to play at best a minor role in trait expression (Fitzpatrick 2004; Fry 2009; Mank 2009). If sex-limited or sex-biased expression is responsible for SD, then one would expect to see significant interactions between the sex-determining region and autosomal loci (e.g., Long et al. 1995; Perry et al. 2003; Delph et al. 2010). Although both mechanisms may be at play during the evolution of SD, the former may be expected to predominate in species with nascent sex chromosomes, whereas the latter may dominate in species with well-developed sex chromosomes (Rice 1996; Charlesworth et al. 2005).

Although information on the genetic basis of SD in animals has steadily been accumulating (e.g., Cowley et al. 1986; Nuzhdin et al. 1997; Prowell 1998; Kopp et al. 2000; Lindholm and Breden 2002; Fitzpatrick 2004; Mank 2009), few studies have addressed this issue in flowering plants (but see Scotti and Delph 2006; Delph et al. 2010) where SD also exists (Delph et al. 1996; Geber et al. 1999; Ashman 2005). None have examined the genetic basis of SD in flowering plant species where separate sexes have evolved relatively recently and SD is considerably less common and less pronounced. The same mechanisms, however, are thought to be involved in generating SD in both plant and animal systems (Delph and Ashman 2006; Bedhomme and Chippendale 2007). Moreover, because the evolution of SD may evolve in concert with the sex chromosomes (Rice 1987), the relatively young age of these chromosomes in some plant species (e.g., <20 million years old [MYO]: Bergero et al. 2008; Yu et al. 2008) presents a unique opportunity to also investigate the timing of the evolution of SD in the context of the evolution of separate sexes (dioecy). One idea is that the accumulation of sexually antagonistic genes that lead to SD occurs after initial suppression of recombination between the two original sex-determining genes (Charlesworth et al. 2005). However, if the initial sterility mutations themselves affect SD, that is, have pleiotropic effects on other primary sexual traits (e.g., “compensatory” mutations, Charlesworth and Charlesworth 1978) or secondary sexual traits (e.g., petal size, Eckhart 1999) then we might see SD prior to the second sterility mutation or suppression of recombination between the two loci. In addition, if genes with sexually antagonistic effects exist on an autosomal chromosome prior to the sterility mutations, then they may facilitate the linkage of the sex-determining gene and the evolution of a sex chromosome (van Doorn and Kirkpatrick 2007; but see Ironside 2010 for an alternate scenario).

A full understanding of how SD evolves also requires an understanding of the genetic relationships between traits. For example, traits not under direct selection may nevertheless become sexually dimorphic if they are positively genetically correlated to traits that are under sexually antagonistic or differential selection (e.g., Ashman 2005; Delph 2007). On the other hand, the same correlations can constrain the evolution of SD if selection is acting on those traits in opposite directions (Conner and Via 1993). Several studies have shown positive genetic correlations between male and female traits such as pollen and ovule production (reviewed in Ashman 2003), which can potentially constrain not only the degree of SD that can be achieved through these traits but also the evolution of completely separate sexes in general. Although a few studies have improved our understanding of the magnitude and prevalence of genetic correlations in plants through classical quantitative genetic approaches (e.g., Meagher 1992, 1999; Ashman 2003; Delph et al. 2004a,b; Steven et al. 2007), QTL analysis can aid in disentangling which and how many regions of the genome are involved as well as the genetic mechanism underlying correlations such as linkage disequilibrium or pleiotropy (e.g., Juenger et al. 2005; Via and Hawthorne 2005), all of which have consequences for predicting the evolution of greater SD. For example, positive correlations between male and female function traits can arise through past selection on alleles at different unlinked loci in hermaphroditic ancestors or through pleiotropy, but only the latter would reflect a constraint for SD.

Fragaria virginiana (Rosaceae), the Virginian wild strawberry, is an exemplary system in which to study the early stages of SD and sex chromosome evolution because it is not fully dioecious, has a proto-sex chromosome, and exhibits SD and quantitative genetic variation for a suite of reproductive and vegetative traits. Using a genetic mapping approach that only addressed sexual functions qualitatively, Spigler et al. (2008, 2010) demonstrated that the three sexes of this subdioecious species (males, females and hermaphrodites) are determined by a proto-sex chromosome containing at least two gene regions controlling male and female fertility that are linked with limited recombination between them. Because the allele conferring male sterility is dominant to the fertility allele (Ahmadi and Bringhurst 1991), females are always the heterogametic sex. Previous quantitative genetic studies have shown differences in genetic (co)variance matrices between the sexes for reproductive traits and genetic correlations that could facilitate or constrain the evolution of SD (Ashman 1999a,b, 2003, 2005). However, a QTL approach to describe the genetic architecture of these has not previously been performed. Such work is particularly novel because F. virginiana is octoploid with disomic inheritance (2n= 8x= 56, Bringhurst 1990; Ashley et al. 2003), and duplicated gene copies might add an additional level of complexity to trait evolution, including that of SD, because the number of possible genic interactions makes polyploids inherently evolutionarily flexible and opportunistic (Wendel 2000; Lawton-Rauh 2003; Adams 2008). Lastly, F. virginiana is a young (∼0.20 MYO, Njuguna 2010) species, and given that sex-limited expression may evolve on longer time scales than sex-linkage because of the regulatory mechanisms required for the former (Rice 1996; Charlesworth et al. 2005), we might expect relatively few epistatic interactions between autosomal loci and the sex-determining region to contribute to SD in F. virginiana.

Here, we use QTL analysis to elucidate the genetic architecture of SD in F. virginiana and to provide insight into how sex chromosomes and SD evolve in general. We begin by asking whether there are regions of the genome other than the proto-sex chromosome that affect female fertility. By treating female fertility as a quantitative trait, we confirm the existence of a single major region housing the sex-determining loci. We then ask the following: (1) Do we find that major QTL for other sexually dimorphic traits are also on the proto-sex chromosome? (2) Are there additional QTL for sexually dimorphic traits on autosomes, and if so, is there an indication of epistatic interactions between these QTL and the sex-determining region that might also contribute to SD? (3) Are genetic correlations that have been previously identified in F. virginiana and are expected from theory on sexual system evolution caused by pleiotropy or linkage disequilibrium?

Methods

MAP CROSS DETAILS

For the QTL analysis, we created an F1 F. virginiana mapping population from a cross between a female maternal parent (Y33b2) and hermaphrodite paternal parent (O477) that were chosen based on test crosses to capture the greatest diversity of putative sex-determining genes (see Spigler et al. 2008). F1 crosses are standard for mapping and QTL studies in highly heterozygous species and/or those that are incapable of or have strong negative effects of inbreeding (Patterson 2002). The mapping parents originated from two wild populations in northwest Pennsylvania that show relatively little neutral genetic differentiation (pairwise FST= 0.02; R. B. Spigler and T-L. Ashman, unpubl. data). We hand-pollinated Y33b2 with pollen collected from O477 and produced three clonal replicates of each of 184 resultant progeny that were maintained in the greenhouse at the University of Pittsburgh in a randomized block design. Each clone was grown in 200-mL pots filled with a 1:2 sand and Fafard no. 4 (Agawam, MA) soil mix, fertilized with 10 beads of Nutricote 13:13:13 N:P:K fertilizer and provided water and pest control as needed throughout the study.

PHENOTYPIC TRAITS

We measured the following reproductive and vegetative traits on each clone to evaluate the level of SD and identify QTL. Male function was assessed based on anther characteristics of four to eight flowers per plant. We scored individuals as “male fertile” when they produced plump, pollen-filled anthers; we scored individuals as “male sterile” when they produced small, white vestigial stamens that lacked pollen-filled anther sacs. When visual scoring of this trait was in question, we inspected excised anther sacs under a compound microscope for the presence of pollen, but in no case did these individuals produce pollen. We confirmed that this score matched across clones of the same genotype. To ensure full potential seed fertilization and fruit production, we hand-pollinated flowers three times per week using outcross pollen. We counted the number of anthers and ovules per flower in two flower buds per plant with known inflorescence positions. We collected three anthers from male-fertile plants and counted pollen using a particle counter (Elzone II 5390 V1.03, Norcross, GA) following Ashman and Hitchens (2000). For four-day-old flowers per plant at either primary or secondary positions, we measured the length and width of a single petal using digital calipers and calculated petal area as the product of petal length and width. We recorded the date of first flower and flowering duration. We estimated female function as the proportion of flowers that became a fruit (proportion fruit set). We calculated proportion seed set from a randomly selected fruit as the number of seeds divided by the total ovules per fruit. If a plant did not make a fruit, we assigned it a value of 0 for proportion seed set. At the end of flowering, we measured the following vegetative traits. We counted the number of leaves, runners, and plantlets. We measured the width of the central leaflet of the largest leaf (leaf size) and lengths of all runners (runner length) and estimated abaxial trichome density from a 0.32-cm2 leaf sample.

For all traits examined, we used the genotypic mean for QTL analysis. In cases where multiple measurements were taken on a trait, we first calculated the average within the clone and then across all three clones to obtain the genotypic mean. Because flower bud position significantly affected ovule and anther number per flower (P < 0.0001), we removed this effect when calculating clone averages using the lsmeans statement in proc glm (SAS software, version 9.1, SAS Institute Inc., Cary, NC).

TESTING FOR SD

For each trait, we tested for significant SD using a t-test (proc ttest in SAS software) or a Wilcoxon rank sum test (proc npar1way in SAS software) and calculated an index of SD (hereafter, “SD index”) following McDaniel (2005) as |(xMSxMF)|/[(SEMS+ SEMF)/2], where x and SE are the mean and standard error, respectively, for each trait for male-sterile (MS) and male-fertile (MF) individuals. We chose this index because it accounts for differences in variance as well as differences between means, and preliminary analyses revealed that this estimate is highly correlated with the index of Lovich and Gibbons (1992) (r= 0.91, P < 0.0001). We applied Bonferroni corrections to the statistical tests for SD.

GENETIC MAP OF F. VIRGINIANA USED FOR QTL

We used the genetic maps of this cross published in Spigler et al. (2010) with the following modifications. We mapped an additional PCR-based marker (BFACT010a) at LOD > 10 to the proto-sex chromosome (linkage group VI-C) in the maternal and paternal maps using JoinMap®4 (Van Ooijen 2006) with the Kosambi mapping function, a minimum LOD threshold of 3.0, recombination threshold of 0.35, and jump threshold of 3.0. Primers for this marker were designed from the full sequence in Genbank (AM889095.1; Rousseau-Gueutin et al. 2008) using Primer-3 (version 0.2; http://frodo.wi.mit.edu/cgi-bin/primer3/primer3_www_slow.cgi). In addition, to be consistently conservative across linkage groups (LGs) for the QTL analyses, we also used the abovementioned parameters to remap two LGs that were mapped previously with less-strict parameters in the paternal map (LGs I-D,VI-C) (see Spigler et al. 2010).

The maps of this outbred population of a polyploid species were constructed using a pseudo-testcross approach (Grattapaglia and Sederoff 1994; Spigler et al. 2010), a strategy that is widely used for highly heterozygous species and/or those for which inbred cross designs are not feasible (e.g., Cervera et al. 2001; Rousseau-Gueutin et al. 2008; Fournier-Level et al. 2009). Because this approach examines each parental meiosis separately, a more efficient way to evaluate QTL is to employ a two-genotype QTL model rather than a four-genotype QTL model by treating our mapping population as a doubled haploid population for QTL analysis (Van Ooijen 2009). This requires removing markers found in single doses in both parents that segregated 3:1 among the progeny (i.e., “hk × hk” segregation types in JoinMap) from the map (Van Ooijen 2009). Finally, we removed the binary phenotypic marker “male sterility” from the map because we did not want to bias positioning of QTL based on its positioning. Regardless, the results are qualitatively similar to those when it is retained (data not shown). We note that the map contains relatively few regions where >2 adjacent markers exhibited skewed segregation ratios (P < 0.01).

QTL ANALYSIS

We performed separate analyses for the parental maps in MapQTL®5 (Van Ooijen 2004). To balance rigor and power given the size of our population (final N= 179), we used a hierarchical approach to identify QTL. We first used a Kruskal–Wallace analysis to identify potential single-marker associations with the phenotypic traits. We subsequently performed interval mapping (IM) analysis using a 1-cM mapping step size to detect QTL. Significance of QTL detected by IM was evaluated at the 5% significance threshold by permutation tests (N= 1000 permutations) at the chromosome level (CL) and genome-wide (GW) level. To minimize false positive QTL at the CL, we used additional criteria to validate QTL found significant at the CL only. These criteria were that (1) a marker adjacent to the QTL was significant at the α= 0.005 level according to single-marker analysis (Van Ooijen 2004) and (2) this marker occurred in a group of markers (≥2) that were significant at α= 0.10 level according to the single-marker analyses when single-marker analysis results were considered in the context of the linkage group map. We proceeded to use composite IM (multiple-QTL model, “MQM”, in MapQTL) only when we detected significant QTL at the GW significance level. To aid in the selection of appropriate cofactors for the MQM analysis, we used the automatic cofactor selection tool in MapQTL with a threshold of P= 0.005. Automatic cofactor and MQM analyses were repeated until a stable set of significant cofactors remained. We calculated 2-LOD intervals for QTL found through MQM and created figures in MapChart (Voorrips 2002).

For those traits for which we found a QTL on the proto-sex chromosome and additional QTL on autosomes, we evaluated autosomal QTL-by-sex interactions using analysis of variance (ANOVA, proc glm in SAS software) (e.g., Delph et al. 2010) because such epistatic interactions may elicit sex-specific effects that lead to SD. We included male function (sterile or fertile) as a fixed effect in the model to represent the QTL that overlap it on the top of LG VI-C (Spigler et al. 2010). We used this approach rather than use the markers most closely associated with the male-function locus (i.e., ARSFL7 or BFACT010a) because male function represents a putative single gene with two alleles, with which we are interested in identifying interactions. In addition, we included as independent variables in the ANOVA markers nearest to the peak of the autosomal QTL for each trait/LG combination identified as significant in IM using the same criteria described above. For cases in which more than one QTL peak was found significant on a given LG according to IM, only the marker associated with the QTL with the highest LOD score was included in the ANOVA model. All markers were treated as binary class variables (1 if present, 0 if absent), and markers from both the maternal and paternal maps were included in the same analysis if applicable. We examined the main effect of male function and each marker as well as all two-way interactions on each phenotypic trait. Note that no data were transformed to be consistent with the QTL analysis.

GENETIC ARCHITECTURE OF TRAIT CORRELATIONS

We examined whether QTL were found on the same linkage group (linked QTL) for a suite of trait pairs shown to be correlated in previous quantitative genetic work (Ashman 1999a,b, 2003). If two traits had a positive genetic correlation then we expected those traits to have linked QTL and their additive effects to be in the same direction. In contrast, if two traits had a negative genetic correlation, then we expected linked QTL but with additive effects in opposite directions. For these comparisons, we considered all QTL that were found to be significant at the CL in IM without the additional significance criteria used above. We also examined whether linked QTL overlapped when in the same parental map. To do this, we estimated 1-LOD intervals for each QTL from the MapQTL output by identifying the positions (cM) flanking the QTL where the LOD score is closest to one minus the peak LOD score. To calculate the probability that QTL are located on the same LG or in the same 1-LOD interval by chance alone, we used the hypergeometric probability distribution function, as follows:

image

where n is the number of intervals (defined here as 30 cM or a LG, depending on the test) that can be compared; m is the number of matches between QTL (i.e., when 1-LOD intervals overlap or are found on the same LG); l is the total number of QTL found for the trait with a greater number of QTL; and s is the total number of QTL found for the trait with fewer QTL (Patterson 2002). We calculated this probability for each parental map separately and, for tests at the level of LGs, across parental maps. If the probability was lower than 0.05, we can conclude that these linked or overlapping QTL reflect underlying genetic processes or patterns rather than chance.

Results

QTL FOR PROPORTION FRUIT SET

QTL analysis of proportion fruit set was consistent with our previous qualitative mapping of female function. Here, we found a single, major QTL for proportion fruit set on the proto-sex chromosome in the maternal parent (LG VI-C) with a LOD score >100 through both IM and MQM (Table 1, Fig. 1, Table S1). This QTL explained approximately 95% of the variation in fruit-setting ability, and its peak was at 2 cM with a narrow 2-LOD interval overlapping the region where male sterility also maps.

Table 1.  QTL results summary showing linkage groups in the maternal and paternal maps where QTL were detected for each trait and highlighting involvement of the proto-sex chromosome. For full detailed data on each QTL see Table S1.
TraitMaternal mapPaternal map
Proto-sex chromosomeAutosomesProto-sex chromosomeAutosomes
  1. Note: Traits are listed in a decreasing order of greatest degree of sexual dimorphism to least (see Table 2). All QTL were found significant at the chromosome level in interval mapping, using additional significance criteria (see Methods). LGs in bold remain significant at the genome-wide level in interval mapping. LGs denoted with “*” are significant at the genome-wide level in MQM mapping.

  2. 1Trait is limited to male-fertile individuals.

Fruit setVI-C*   
Petal areaVI-C*III-A, VI-B  
Anther numberVI-C*III-B, VI-D VI-A
Seed setVI-C*III-C*, V-B VI-A (2-p)
Leaf numberVI-C*II-C, V-BVI-C 
Flowering durationVI-C*III-C, VII-A II-A
Runner lengthVI-C* VI-C*III-A*, V-D, 4-p
Flower numberVI-CI-A, IV-C, V-B, VI-B II-A*, V-B
Date of first flowerVI-CI-B, II-D, III-C*, III-D, V-D, VI-B, VII-B, VII-C V-B
Leaf size VI-B  
Runner number V-B, VII-A II-C, VII-A
Trichome number II-B* II-A, III-D
Ovule number III-D* I-C, II-D, IV-A*,  VI-A*, VI-B*
Plantlet number VII-D III-A, III-C, VI-D, VII-A
Pollen per anther1 II-B  
Figure 1.

Location of QTL for sexually dimorphic traits on the proto-sex chromosome (VI-C) in the maternal parent (m) and paternal parent (p). QTL were detected using MQM mapping, and 2-LOD intervals are provided for each QTL. The location of the qualitative trait marker “Male Sterility” is indicated on the maternal proto-sex chromosome for reference.

SEXUAL DIMORPHISM

Most traits examined were significantly different between male-fertile and male-sterile individuals (Table 2). However, the degree of SD varied widely, ranging from 0.89 to 54.29. Proportion fruit set and petal area were the most dimorphic, whereas ovule number, trichome density, and plantlet number had the lowest SD indexes and were also statistically indistinguishable from monomorphism (Table 2).

Table 2.  Mean trait and standard error values for male-sterile and male-fertile individuals in the F. virginiana mapping population and tests for and estimates of the degree of sexual dimorphism.
TraitF1 progeny means (SE) Test for SDSD Index2
Male-sterileMale-fertilet or Z1P
  1. Note: Traits are sorted from highest to lowest degree of sexual dimorphism as calculated by the SD Index. P-values highlighted in bold remain significant after Bonferroni correction.

  2. 1t or Z scores are provided, as indicated by superscripts (t and Z), depending on whether data were evaluated using a t-test or Wilcoxon rank sum test, respectively.

  3. 2SD Index=|(xMSxMF)|/[(SEMS+SEMF)/2], where x and SE are the mean and standard error, respectively, for each trait for male-sterile (MS) and male-fertile (MF) individuals (following McDaniel 2005).

Fruit set 0.88 (0.015)  0.09 (0.014) 10.90Z<0.000154.29
Petal area47.67 (1.346) 92.26 (1.278)−11.39Z<0.000133.98
Anther number20.52 (0.179) 22.89 (0.17) −9.60t<0.000113.58
Seed set 0.71 (0.028)  0.44 (0.026)  5.71Z<0.0001 9.82
Leaf number10.86 (0.254) 13.17 (0.241) −6.60t<0.0001 9.34
Flowering duration33.43 (0.728) 38.36 (0.692) −4.91t<0.0001 6.95
Runner length85.65 (2.863)100.98 (2.749) −3.86t 0.0002 5.46
Flower number25.14 (1.033) 30.46 (0.982) −3.32t 0.0009 5.28
Date of first flower59.85 (0.574) 57.35 (0.545)  3.18Z 0.0015 4.48
Leaf size36.48 (0.462) 34.87 (0.439)  2.53t 0.0124 3.57
Runner number 1.83 (0.076)  1.61 (0.072)  2.14Z 0.0321 2.97
Trichome density24.24 (1.021) 25.85 (0.975) −0.93Z 0.3549 1.62
Ovule number82.74 (1.394) 80.73 (1.324)  1.05t 0.2969 1.48
Plantlet number 4.14 (0.195)  4.31 (0.187) −0.47Z 0.6381 0.89

GENETIC ARCHITECTURE OF SEXUALLY DIMORPHIC VERSUS MONOMORPHIC TRAITS

For almost all (8 out of 10) sexually dimorphic traits, we detected QTL through IM at the top of the proto-sex chromosome (VI-C) linked to the region where the dominant male sterility allele and the QTL for proportion fruit set are located in the maternal map (Table 1, Table S1). The proximity of the peaks of these QTL (most within 2 cM) to the male function locus suggests either pleiotropy or close linkage of multiple genes. This result was further confirmed for seven traits through MQM analysis, revealing 2-LOD intervals that overlap this region (Fig. 1, Table 1, Table S1). We also detected major QTL on the proto-sex chromosome in the paternal map for two sexually dimorphic traits (leaf number and runner length) through IM, and the QTL for runner number was significant at the GW level (Fig. 1, Table 1, Table S1). However, not all sexually dimorphic traits had QTL on the proto-sex chromosome; we only detected QTL for runner number and leaf size on autosomal LGs (Table 1, Table S1). We detected only a single QTL for pollen per anther, a trait limited to male-fertile individuals, through IM, and it was not located on the proto-sex chromosome (Table 1, Table S1). For the three sexually monomorphic traits we examined, we found multiple QTL (3–6 using IM; 0–4 using MQM) located solely on autosomes across both parental maps (Table 1, Table S1).

For eight of the sexually dimorphic traits, we detected between two and eight autosomal QTL across both parental maps through IM in addition to the QTL on the proto-sex chromosome (Table 1). Therefore, we proceeded to evaluate potential epistatic interactions between autosomal QTL and the male-function locus using ANOVA for these eight traits. Only for proportion seed set was the interaction between male function and autosomal markers identifying QTL significant (LG VI-A, marker CO816667_ 316: F1= 7.85, P= 0.006; LG III-C, marker EMFn202_251: F1= 4.67, P= 0.033), indicating that the effect of these autosomal QTL depends upon the presence of the male-sterility allele. For example, marker CO816667 associated with the QTL on LG VI-A in the paternal parent reduced proportion seed set in male-fertile individuals by approximately 20% (Tukey's post-hoc test P= 0.002) but had no effect in male-sterile individuals (P= 1.0) (Fig. 2). Marker EMFn202 in the maternal parent had a similar effect, illustrating the sex-specificity of these QTL (Fig. 2). Together, these QTL and their interactions explain 44% of the variation in proportion seed set. Although no other significant interactions were found (P > 0.05, data not shown), these ANOVAs confirmed the main effects of the autosomal QTL included in the models (P < 0.05) in almost all cases, indicating that additional QTL outside the sex region influence many of the SD traits examined. For proportion seed set and runner length these results were further confirmed by MQM analysis, again indicating genetic control of variation in those traits that is at least partially independent from the sex-determining region (Table 1, Table S1).

Figure 2.

Effect of the interaction between two SSR markers (CO816667_316 and EMFn202_251) and the male-function locus on proportion seed set. Both SSR markers significantly reduce proportion seed set in male-fertile individuals, but not male-sterile individuals. On the x-axis, “0” indicates the absence of each marker, and “1” indicates presence. Note that CO816667_316 is inherited from the paternal parent; EMFn202_251 is inherited from the maternal parent. Mean proportion seed set and standard error bars are given, as determined using the least squares mean statement for the interaction terms in the ANOVA.

GENETIC ARCHITECTURE OF TRAIT CORRELATIONS

When we looked across the 30 LGs in both parental maps, for eight of nine pairs of traits examined we found a total of 23 cases of linked QTL (Table 3). For one pair of traits (anther number and petal area), QTL were found on the same LG more often than expected by chance (hypergeometric probability: P= 0.03). When we looked at these patterns within each parental map, 17 of the 23 linked QTL were found within a given parental map, allowing comparison of QTL placement along LGs. In the majority of cases (16 of 17), the linked QTL had overlapping 1-LOD intervals, and in eight of those cases, QTL peaks were within 2 cM of one another. These observed patterns of overlap are statistically significant for several pairs of traits: proportion fruit set and petal area QTL in both maps (Pmaternal= 0.006, Ppaternal= 0.04), proportion fruit set and anther number QTL (Pmaternal= 0.017), anther number and ovule number QTL (Ppaternal= 0.016), and anther number and petal area QTL (Pmaternal= 0.01).

Table 3.  Genetic correlations between traits as revealed by linked QTL and their fit to expected relationships. Linkage groups where linked QTL were found are given.
 Fruit setFlower numberOvule numberAnther number
  1. Note: Plus and minus signs in parentheses indicate the expected relationship between QTL for traits of interest based on previous work (Ashman 1999a,b, 2003, 2005). Linkage groups (LG) in bold have QTL with additive effects for each trait that agrees with the expected sign of the correlation. Italics indicate that the signs of the additive effects of the QTL on the LG are in opposition to the predicted relationship. Underlined font indicates cases where 1-LOD intervals for the QTL overlap; an asterisk (*) is shown where QTL are within 2 cM of each other. Regular font indicates cases in which QTL were found on the same LG but in different parental maps, precluding meaningful comparisons of additive effects and positioning. Note that the trait “pollen grains per anther” was not included in the table because no QTL were shared with any other traits. We also note that these correlations reflect variation in traits both within and between sexes, but the direction was expected to be the same based on previous work.

  2. 1Previous work has demonstrated only a phenotypic correlation between these traits; therefore, no shared or linked genetic basis is expected.

  3. 2QTL were detected on LG 4-m, which was determined to be the top of LG VI-D in Spigler et al. (2010).

  4. 3LG VI-C is listed twice because a set of linked QTL for these traits was found in both parental maps. QTL are within 2 cM of each other in the maternal map.

Flower number(0)1 n/a      
Ovule number(+) none(+)IV-C    
    VI-A    
    VI-B    
Anther number(−) VI-C*(−)VI-A(+)VI-A  
  VI-D (4-m)2 VI-B VI-B*  
    VI-C*    
Petal area(−) III-A(+)VI-A(+)VI-A(+)VI-A*
  VI-C*3 VI-B VI-B VI-B
  VI-C VI-C*   VI-C*
  VI-D (4-m)*     VI-D (4-m)

Furthermore, for most trait pairs for which we could compare signs of the additive effects (five out of seven), the signs of linked QTL fit predictions from previous work. For instance, consistent with a trade-off between male and female function, linked QTL for proportion fruit set and anther number and for proportion fruit set and petal area had opposing effects. Positive correlations between flower number and petal area and between petal area and anther number were also confirmed; these pairs of traits had at least two sets of linked QTL with additive effects in the same direction. Results for ovule number and anther number also support previous findings of positive genetic correlations between male and female function within flowers. There were cases where the direction of additive effects of linked QTL did not match predictions. One limitation to these tests, however, is that we cannot compare the additive effects of linked QTL that are on the same linkage group but in different parents. Another noteworthy finding is that the majority of cases (20 out of 23) of linked QTL involved the proto-sex chromosome (VI-C) and other LGs from the same homeologous group (HG), namely VI-A, VI-B, and VI-D. Moreover, all of the HG VI LGs housed QTL influencing more than two of the traits of interest. For example, VI-A and VI-B have QTL that influence petal size, anther number, ovule number, and flower number.

Discussion

This study shows that both sex-linked and sex-limited QTL contribute to SD in a subdioecious plant. The predominance of sex-linkage, however, is consistent with predictions for nascent sex chromosomes. Moreover, we also detect QTL on autosomes for sexually dimorphic traits, demonstrating that regions outside the proto-sex chromosome contribute to phenotypic variation in these traits. Additionally, our work characterizes the genetic architecture of several important correlations between reproductive traits as evidenced by the existence of linked QTL, and these reveal the potential importance of HG VI LGs in reproductive trait variation. Our results are likely representative of F. virginiana in general because our map parents were selected from the wild, and thus the mapping population reflects the genetic variation and levels of SD seen in wild populations (Ashman 1999a,b, 2003, 2005). Below we discuss our main findings in the context of present knowledge.

THE PROTO-SEX CHROMOSOME IN F. VIRGINIANA

We show that quantitative variation in proportion fruit set is controlled almost exclusively by a single region linked to the male-function locus on LG VI-C, confirming the presence of a proto-sex chromosome in F. virginiana (Spigler et al. 2008, 2010). This finding represents a significant advance over previous work that treated female function as a qualitative marker and thus constrained it to map to a single location (Spigler et al. 2008, 2010). Although estimation of the percent variation explained by this (and other) QTL in our study is subject to overestimation because of the Beavis effect (Beavis 1994, 1998), the only other QTL for proportion fruit set was detectable when we relaxed our significance criteria still only explained a minor fraction of the variation (<5%). We acknowledge that a larger study would provide greater power to detect minor QTL if they exist and/or to evaluate those that contribute to within-sex variation in proportion fruit set. Additional crosses could address the question of whether populations vary in the number of QTL contributing to fruit set.

THE ROLE OF THE PROTO-SEX CHROMOSOME IN SD

The sex-determining region of F. virginiana significantly influences variation in almost all traits we studied, thus causing SD. Similarly, Delph et al. (2010) recently demonstrated QTL for 40 sexually dimorphic traits overlapping the sex-determining region in Silene latifolia, a fully dioecious species with heteromorphic and older sex chromosomes. In addition, the sex-determining locus in grape was shown to have QTL for ovary and stamen length (Marguerit et al. 2009). Although we also detected multiple QTL on autosomes that contributed to variation in sexually dimorphic traits, a strong correlation between the degree of SD and the percent variation explained by the QTL linked to the proto-sex chromosome (r= 0.97, P < 0.0001) is consistent with predictions from sex-linkage as the primary cause of SD (Fairbairn and Roff 2006). But, we do not yet know the precise genetic mechanism underlying the sex-linked QTL. It is worth noting that the initial sterility mutations themselves (here, dominant male sterility) are predicted to have negative pleiotropic effects on sex function (Charlesworth and Charlesworth 1978), and we observed overlapping QTL with opposing effects on proportion fruit set and anther number. Of course, the sex-determining region in F. virginiana, as currently delineated by at least two loci that control male and female function (Spigler et al. 2008, 2010), may house as many as 150 genes (based on Pontaroli et al. 2009 and Spigler et al. 2010). Thus there is ample opportunity for multiple genes to be contributing to the overlapping QTL in this region. One way to test this would be to remove the influence of the male sterility locus by estimating QTL with only one sex in a larger mapping population (e.g., Scotti and Delph 2006) and examine whether we still find QTL in the same region. In addition, identifying candidate genes underlying these QTL will aid in determining whether these QTL represent multiple closely linked loci.

Although theory suggests that linkage to the sex-determining region is the predominant cause of SD, SD can also arise via epistatic interactions between the sex-determining region and autosomal loci that result in sex-limited or sex-biased gene expression (Mank 2009). We were able to detect multiple autosomal QTL for sexually dimorphic traits and for two of these QTL revealed epistatic interactions with the male function locus for one trait (proportion seed set) (Fig. 2). In the only other plant study available for comparison, Delph et al. (2010) were able to detect sex-limited QTL for 20% of SD traits in S. latifolia with a similar analytical approach. This difference could reflect interesting biological differences, such as the age of the sex chromosomes, because sex-limited loci are thought to evolve on longer time scales than sex linkage (Rice 1996; Charlesworth et al. 2005). Alternatively, it is possible that other sex-limited QTL exist in our mapping population, but we were not in a position to detect them because of the constraints imposed by an F1 cross, small population size or potential monomorphism at autosomal loci.

SD: BEFORE OR AFTER?

The association between the sex-determining region and multiple QTL begs the question as to when this association, and thus SD, arose in F. virginiana. However, because the evolution of SD and sex chromosomes is a cyclical process—once recombination between the sex-determining genes is suppressed, sexually antagonistic genes may accumulate, which, in turn, may select for increased recombination suppression, leading to sex chromosome differentiation (Bull 1983; Rice 1987; Charlesworth et al. 2005)—we cannot answer definitively. Nonetheless, we consider our data in light of other evidence relevant to this process that has been published and with particular emphasis in Fragaria. First, although the reproductive and vegetative traits examined in our study and, more generally, in plants may not classically be thought of as antagonistic, we know that selection on floral traits via female fertility opposes that via male fertility in several species, including F. virginiana (reviewed by Ashman and Morgan 2004; Ashman 2005; Delph and Ashman 2006, T-L. Ashman unpubl. data), thus it is possible that sex-differential selection may have been involved in the association between the sex-determining gene(s) and QTL for the sexually dimorphic traits examined in the current study. Second, because the presence of individuals that are both male-sterile and female-sterile (neuters) in F. virginiana indicates recombination between the male-function and female-function loci (Spigler et al. 2008), our results suggest that accumulation of sexually antagonistic genes may occur even before recombination between the original sex-determining loci is fully suppressed. Although this accumulation is classically discussed as occurring after initial recombination suppression, experimental work in Drosophila demonstrated the rapidity with which a new sex-determining locus can accrue linked sexually antagonistic genes (Rice 1992), supporting the hypothesis that this accumulation can occur in species without dioecy such as F. virginiana.

Alternatively, the genes for traits that became sexually dimorphic may have resided on LG VI-C prior to it acquiring the sex-determining genes. Indeed, reproductive (including flower size and self-incompatibility) and vegetative (leaf size and runner length) traits have been mapped to the homoeolog of the proto-sex chromosome (VI-C), that is LG 6 in a diploid hermaphrodite Fragaria cross (Sargent et al. 2004, 2007; Sargent 2005; Bošković et al. 2010). These findings as well as our evidence of the disproportionate number of linked or pleiotropic reproductive QTL housed on homeologous LGs in HG IV suggest that at least some of the genes underlying the QTL detected on LG VI-C in F. virginiana existed prior to polyploidization and possibly prior to the acquisition of the sterility alleles and that they are distinct from the locus controlling male function. These findings have wide implications, as van Doorn and Kirkpatrick (2007) proposed that autosomal genes under antagonistic selection may facilitate linkage of new sex-determining genes or the movement of already established ones, thus predisposing certain autosomes to become sex chromosomes (see also Graves and Peichel 2010). This scenario could occur even in the absence of sexually antagonistic selection, if combinations of favorable alleles existed alongside the sex-determining genes in their ancestral state, that is, prior to the acquisition of the sterility mutations, which could also lead to recombination suppression on sex chromosomes (Ironside 2010).

GENETIC ARCHITECTURE OF SEXUAL TRAIT CORRELATIONS FACILITATES THE EVOLUTION OF SD

Our results go beyond previous quantitative genetic work in F. virginiana (Ashman 1999a,b, 2003, 2005) by showing that genetic correlations between reproductive traits are due to linkage on the same chromosome and in several cases represent overlapping QTL suggesting plieotropy or close linkage. Our confirmation of linked QTL for a male–female trade-off, specifically between anther number and proportion fruit set, is a first for this essential assumption of sexual system evolution theory (e.g., Charnov 1982; Charlesworth 1999). In addition, our work illustrates the complexity of the genetic architecture underlying genetic correlations. For almost every pair of traits we examined, we found at least two regions of the genome that contributed to their overall genetic correlation. Most cases of linked QTL that we detected among traits can allow the evolution of separate sexes and SD, for example, positive integration of male function traits or opposing effects of linked QTL underlying fruit set and petal area, whereas our corroboration of a positive genetic correlation between within-flower male–female function traits (Ashman 2003), specifically anther and ovule number, suggests a possible constraint. Additional studies, including multitrait QTL analysis (Jiang and Zeng 1995) and multiple generations of enforced recombination (Conner 2002), can help to further distinguish whether the overlapping QTL found in our study represent close linkages that can be broken or are due to pleiotropy.

CONCLUSIONS AND FUTURE DIRECTIONS

Our study is one of the first to examine the genetic architecture of SD in a plant species using QTL analysis and the first in a species without complete dioecy. We clearly demonstrate that the sex-determining region also controls significant variation in important sexual, vegetative, and phenological traits. Although comparison of our results to the small set of studies of its kind in plants can be informative (Marguerit et al. 2009; Delph et al. 2010), consideration of our results in combination with those from diploid hermaphroditic Fragaria (Sargent et al. 2004, 2007; Sargent 2005; Bošković et al. 2010; see above) can reveal perhaps even more about the nature of sex chromosomes and the evolution of SD. In particular, we suggest that chromosome 6 may have been a hub of linked and/or pleiotropic reproductive genes prior to polyploidization and the acquisition of the male and female sterility mutations. If these traits were also under antagonistic selection or were linked such that they acted as a “supergene” (Ironside 2010) via male and female fitness components in the ancestral hermaphrodite species (e.g., Ashman and Morgan 2004; Delph and Ashman 2006), then this chromosome may have been predisposed to become a sex chromosome. A next step is to test the possibility that antagonistic selection predates acquisition of sex-determining loci by conducting phenotypic selection analyses in hermaphroditic Fragaria species. At the same time, further characterization of the sex chromosomes of the dioecious sister species F. chiloensis (Goldberg et al. 2010) and other dimorphic congeners can provide a comparative framework that, together with results from the diploid hermaphroditic species, could illuminate the full sequence of the evolution of SD and sex chromosomes within a single genus. Alternatively, such studies might reveal multiple independent origins of SD and sex chromosomes, highlighting the dynamic nature of this evolutionary process, as has been seen in systems such as Silene and fish (Takehana et al. 2007; Mrackova et al. 2008; Ross et al. 2009), and the potential role of polyploidy. Regardless, more studies are needed in young, dioecious or subdioecious species so that we can understand the earliest stages of evolution of sex chromosomes and the origin of SD.


Associate Editor: J. Pannell

ACKNOWLEDGMENTS

We thank D. Cole, C. Collin, E. Early, J. Enns, M.T. Goldberg, S. Goode, E. Korns, A. Johnson, B. McTeague, E. Poor, J. Robinson, H. Tamm, L. Wright, and E York for greenhouse and laboratory assistance. A. Liston, Q. Song, J. Pannell, and anonymous reviewers provided helpful comments on a previous version of the manuscript. J. van Ooijen kindly provided helpful advice on using MapQTL. This work was supported by the University of Pittsburgh, USDA-ARS, NSF (DEB 0449488 and 1020523) to TLA and USDA-CSREES (2005-00765) to TLA and KL. Mention of trade names or commercial products in this article is solely for the purpose of providing specific information and does not imply recommendation or endorsement by the United States Department of Agriculture.

Ancillary