SEARCH

SEARCH BY CITATION

Keywords:

  • ABC model;
  • association analysis;
  • candidate genes;
  • petal and stamen length;
  • stamen exsertion

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information
  • The genetic architecture of floral traits is evolutionarily important due to the fitness consequences of quantitative variation in floral morphology. Yet, little is known about the genes underlying these traits in natural populations. Using Arabidopsis thaliana, we examine molecular variation at GIBBERELLIC ACID REQUIRING 1 (GA1) and test for associations with floral morphology.
  • We examined full-length sequence in 32 accessions and describe two haplotypes (comprising four nonsynonymous polymorphisms) in GA1 that segregate at intermediate frequencies. In 133 A. thaliana accessions, we test for genotype–phenotype associations and corroborate these findings in segregating progenies.
  • The two common GA1 haplotypes were associated with the length of petals, stamens, and to a lesser extent style-stigma length. Associations were confirmed in a segregating progeny developed from 19 accessions. We find analogous results in recombinant inbred lines of the Bayreuth × Shahdara cross, which differ only at one of 4 SNPs, suggesting that this SNP may contribute to the observed association.
  • Assuming GA1 causally affects floral organ size, it is interesting that adjacent petal and stamen whorls are most strongly affected. This pattern suggests that GA1 could contribute to the greater strength of petal–stamen correlations relative to other floral-length correlations observed in some Brassicaceous species.

Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Significant progress has been made in identifying the genes underlying qualitative variation in floral-organ identity among widely separated clades of flowering plants (e.g. the Floral Genome Project; Kim et al., 2005; Theissen & Melzer, 2007). Yet, among natural populations of the same or closely related species pairs, the majority of segregating phenotypic variation is quantitative rather than qualitative (Mackay, 2001). Interspecific crosses have been used successfully to characterize the QTL architecture of differences in nectar production, floral organ coloration, floral shape, and floral organ size in a variety of congeneric species pairs (Bradshaw et al., 1995; Bradshaw & Schemske, 2003; Stuurman et al., 2004; Goodwillie et al., 2006; Bouck et al., 2007; Moyle, 2007). Only a handful of studies, however, have mapped floral morphological QTL segregating between populations within a single species (Juenger et al., 2000, 2005; Hall et al., 2006; Shomura et al., 2008; Hao & Lin, 2010). Furthermore, few QTL contributing to inter- or intraspecific floral differences have been cloned (floral-pigment pathways, Zufall & Rausher, 2004; but see also floral-scent pathways, Verdonk et al., 2005). Therefore, little is known about the identity of genes underlying quantitative variation in floral traits. This lack of genetic characterization stands in pronounced contrast to the thorough evolutionary characterization of floral traits; that is, the hundreds of evolutionary studies that have shown either an association or causal effect of floral organ size and shape on reproductive success (Jain, 1976; Grant, 1994; Ashman & Morgan, 2004; Harder & Johnson, 2009). Here, we take a candidate-gene approach, and identify one gene associated with natural variation in the size of multiple floral organs; this association is further examined through segregation analysis in two synthetic segregating progenies.

Through the use of experimental genetic approaches such as mutant screens and transgenic overexpression, molecular biologists are rapidly characterizing the genes that act in diverse developmental pathways (Simpson & Dean, 2002; Krizek & Fletcher, 2005; Franklin, 2008). This rapid and ongoing annotation of developmental loci can be used to formulate a list of candidate loci that contribute to phenotypic variation in natural populations. Perhaps due to the small size of flowers in the model plant species, Arabidopsis thaliana, comparatively few genes have been identified that affect the size of floral organs. Nevertheless, cell division and elongation genes emerge as logical candidates for loci influencing floral morphological variation, as do genes acting in hormone pathways (Mizukami, 2001; Weiss et al., 2005), and several mutants involved in either cell elongation/proliferation (Clark et al., 1997; Krizek, 1999; Carles et al., 2004) or in hormone production/regulation (Koornneef & Vanderveen, 1980; Goto & Pharis, 1999; Hu et al., 2003) affect floral organ size.

With regard to natural variants, Juenger et al. (2000) mapped 18 QTL for floral morphology using segregating progenies (recombinant inbred lines; RILs) of Arabidopsis, and several annotated candidate genes affecting organ size and development localized to significant QTL. For example, a QTL on chromosome 1 accounted for significant variation in stamen length. One candidate gene within this QTL region, UNUSUAL FLORAL ORGANS (UFO), is an F-box protein that interacts with LEAFY (LFY) to promote APETALA3 (AP3) a floral organ identity gene, which, in turn, regulates petal and stamen development (Chae et al., 2008). Modified expression of UFO (through knock-out mutants or fusion to a 35S promoter) can result in altered number and size of floral organs (Lee et al., 1997; Chae et al., 2008), and members of the APETALA gene family can affect floral organ identity and size (Jofuku et al., 1994; Okamuro et al., 1996). It is worth noting that hundreds of genetic loci are harboured within the confidence limits of most QTL (Ungerer et al., 2003). Therefore, aside from genome-wide QTL screens that can require additional extensive recombination-based fine-scale mapping, candidate-gene analyses with additional co-segregation studies provide a starting point for determination of causal loci.

Several alternative genetic models could account for floral organ size and allometry. Some genes may influence floral organs in all whorls (Goto & Pharis, 1999), because all floral organs must elongate to some degree during development. However, other genes may only affect a specific floral whorl (e.g. petal size; Crawford et al., 2004; Szecsi et al., 2006) or floral-trait pairs (Weigel & Meyerowitz, 1994). Correlated shifts in floral organ sizes may arise in part because a pleiotropic gene affects variation in several traits (Lande, 1980), as in the ABC model of floral development, where B-class genes affect the development of both petals and stamens (Weigel & Meyerowitz, 1994). Analogous to the ABC model for organ identity, one might also postulate a model for organ size under which correlations are stronger between organs residing in adjacent floral whorls than organs in nonadjacent whorls. In prior studies of floral genetic architecture, the size of floral organs is indeed highly correlated such that organs from all whorls are equally strongly associated in some species (Carr & Fenster, 1994; Juenger et al., 2000; Brock & Weinig, 2007). In other cases, pairs of organs from adjacent whorls are more strongly correlated than those in nonadjacent whorls, for example, genetic correlations between filament and corolla length are significantly stronger than all other floral–floral correlations in several Brassicaceous species (Conner & Sterling, 1995; Conner et al., 2009). QTL mapping and analysis of segregating progenies with extensive generations of recombination suggests that genes with pleiotropic effects (or loci in tight physically linkage) are primarily responsible for floral trait correlations in some species (Juenger et al., 2000; Conner, 2002), although in natural populations the role of pollinator-mediated selection and ensuing linkage disequilibrium (LD) of physically unlinked genes should not be discounted for other cases. Regardless of the relative importance of pleiotropy vs. LD, the identity of the underlying loci remains largely unknown.

Here, we use a candidate-gene approach in Athaliana to test the effects of one gene, GIBBERELLIC ACID REQUIRING 1 (GA1) on floral morphology. This locus catalyses the first step in the biosynthetic pathway of gibberellin (GA) production and its promoter is highly active in rapidly elongating plant organs, including roots, leaves, flowers and seeds (Silverstone et al., 1997). Gibberellins have been shown to promote the late-stage expression of several B- and C-class genes in the ABC model of floral development (AP3, PISTILLATA (PI), and AGAMOUS (AG)) (Coen & Meyerowitz, 1991; Yu et al., 2004); thus, the ultimate product of GA1 is necessary for appropriate floral development. In support of its role as a candidate floral development gene, ga1 null mutants show severely underdeveloped petals, stamens and, to a lesser degree, sepals and pistils; additionally, treatment of ga1 mutants with bioactive gibberellin recovers the wild-type phenotype (Koornneef & Vanderveen, 1980; Goto & Pharis, 1999). However, the role of natural variation at this locus in intraspecific variation in floral organ size is unknown. Examination of publicly available sequence data reveals several nonsynonymous polymorphisms within this gene. Using a candidate-gene association analysis in A. thaliana, we test the correlation between two common haplotypes (as well as four SNPs that define them) at GA1 and the size of petals, stamens and pistils. Gene trees indicate that these two haplotypes are well resolved and form divergent basal branches. We observe a significant association between GA1 and floral organs in all whorls of the flower. This association is also observed in a set of Multiparent Advanced Generation Intercross (MAGIC) lines and in a second population of segregating progeny, the Bay-0 × Sha RILs. Although we cannot discount the possibility that GA1 is in LD with a physically proximate causal locus, the combined experimental results strongly suggest a causal effect of natural variation in GA1 on floral organ size.

Materials and Methods

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Plant materials

Arabidopsis thaliana (L.) Heynh (Brassicaceae) is a small, selfing, annual plant that is native to Eurasia but has become naturalized worldwide (Al-Shehbaz & O’Kane, 2002). Small white flowers are produced on both primary and basal inflorescences, which arise from a rosette of leaves. As typified by the family, each flower consists of four sepals, four separate petals, six stamens (two short + four long) and a compound pistil with two fused carpels. Floral size and shape can differ dramatically among accessions of A. thaliana (Juenger et al., 2000; Brock et al., 2009); however, within a plant, flowers from nodes 4–12 have been shown to be developmentally equivalent (Diggle, 1997).

Molecular variation in GA1

We obtained complete GA1 sequences from 20 A. thaliana accessions and one accession of Arabidopsis lyrata from NCBI (for GenBank numbers see Supporting Information Table S1), and an additional 12 sequences from MAGIC lines parental accessions (http://plants.ensembl.org/Arabidopsis_thaliana/Info/Index). We aligned these sequences using Clustal W (Larkin et al., 2007) and removed regions with missing data by deleting indels in introns, producing a 4423 bp fragment (2415 bp coding and 2008 bp noncoding). Sequence data for a 49 bp coding region (codon positions 2199–2247) was missing in the Cvi accession but monomorphic in the remaining individuals; we removed this region from the analysis. To explore the phylogenetic relationship among sequences we constructed gene trees using Bayesian Inference (MrBayes v3.1.2; Ronquist & Huelsenbeck, 2003). For the analysis, we created four unlinked partitions across the GA1 region (1st, 2nd and 3rd codon positions, and the intron regions). For each partition, we used a general time reversible (GTR) nucleotide substitution model (Yang, 1994) with six rate categories and a gamma distribution for variable sites and a proportion of the invariant sites. In a Bayesian framework, this model is robust to over-parameterization of substitution rates (Ronquist & Huelsenbeck, 2003); however, because the second codon position partition did not have substitutions of all types (e.g. CG and TG mutations did not occur in GA1 second codon position), we confirmed initial results by running additional, simpler models (HKY, Kimura, 1980; and F81, Felsenstein, 1981) for the second partition as suggested (Ronquist & Huelsenbeck, 2003). Gene tree topology did not change across these various model comparisons, and we report the tree based on initial settings. Posterior probabilities of the gene tree nodes were estimated using a four Markov chain (one cold and three heated chains with Metropolis-coupling) Monte Carlo analysis for five million generations with a 25% burn-in period, sampling every 50th generation.

In order to identify structural variation associated with the tree topology that might associate with floral morphology, we identified nonsynonymous variation in the predicted amino acid sequence. Using this population of 32 accessions plus additional SNP data available at The Arabidopsis Information Resource (TAIR), we selected nonsynonymous polymorphisms with a minimum frequency of 25%, so that each SNP (and their combined haplotypes) would have sufficient replication to test for genotype–phenotype associations. Based on these criteria we selected the following four nonsynonymous polymorphisms: GA1_467 (number refers to the coding-region nucleotide position), GA1_1169, GA1_2040 and GA1_2354 (Table 1a; at TAIR these variants are named as BKN000007942, PERL0670372, ossowski_668474 and PERL0670355, respectively; Supporting Information Fig. S1a shows locations of exon SNPs). Based on amino acid substitution matrices (Henikoff & Henikoff, 1992), two substitutions (GA1_467 and GA1_2040) are predicted to be favoured substitutions (based on physical or chemical properties) while the other SNPs (GA1_1169 and GA1_2354) result in substitutions that are predicted to be disfavoured (based on physical or chemical properties; Blake & Cohen, 2001). Additionally, the putative disfavoured substitution between glutamine and leucine at GA1_1169 occurs eight amino acids downstream of the aspartate-rich box ‘DIDDTA’, a highly conserved region involved in substrate binding (Sun & Kamiya, 1994; Ait-Ali et al., 1997; Sun & Kamiya, 1997; Koeksal et al., 2011). The presence of putative disfavoured substitutions that segregate at intermediate frequencies, one of which neighbours a putative functional region of GA1, in concert with previous studies implicating a role of GA1 in floral development (Coen & Meyerowitz, 1991; Goto & Pharis, 1999; Yu et al., 2004), suggests that natural variation in GA1 may be associated with variation in floral morphology.

Table 1.   Genetic variation in GA1
(a) GA1 SNPsSNP of the coding strandAmino acid variation
Col-0 (freq.)Variant (freq.)Col-0 AAVariant AA
GA1_467A (0.68)T (0.32)Tyr; YPhe; F
GA1_1169A (0.50)T (0.50)Gln; QLeu; L
GA1_2040G (0.41)T (0.59)Met; MLeu; L
GA1_2354T (0.62)C (0.38)Leu; LSer; S
(b) Haplotype sequenceHaplotype banding patternNFrequency (%)Haplotype name
  1. (a) Four nonsynonymous polymorphisms and associated predicted amino acid in the GA1 coding region and the frequency of each SNP variant in a large panel of Arabidopsis thaliana accessions (N = 133). (b) Haplotype sequences of these four SNPs, banding patterns from CAP genotyping (c, cut; n, not cut), and the haplotype frequency in the Arabidopsis panel.

AAGTcccn4836.1GA1A
TTTCnnnc3828.6GA1B
ATTCcnnc2619.5 
AATCccnc118.3 
AAGCcccc53.8 
TATCncnc32.3 
TTGTnncn21.5 

In order to assign accessions to genotypic states at each of the four SNPs, we developed cleaved amplified polymorphic sequences (CAPs) or derived CAPs (dCAPs; Neff et al., 1998) for all four sites (for primers see Table S2). We genotyped 133 A. thaliana accessions at each SNP (see Table S3; ‘rarest’ SNP state frequency range in this population; 0.32–0.50). Two SNP haplotypes (GA1A vs GA1B; Table 1b) segregate at intermediate frequencies and occupy divergent basal branches of the gene tree (Fig. S1b); as a result, we have sufficient replication of these two distinct haplotypes, as well as individual SNPs in the full panel of Athaliana accessions, to test for associations with floral morphological traits.

Association mapping in accessions: phenotypic measurements of floral morphology

We measured floral morphology of A. thaliana accessions in a glasshouse experiment that included manipulations of light quality at the University of Minnesota (St. Paul, MN, USA). Floral morphology has been shown to vary with the ratio of red to far-red light (R : FR) (Brock & Weinig, 2007). This light quality is a predictor of local plant density and is the abiotic cue for the foliar-shade response in plants, which is characterized by elongation of internodes and petioles, and accelerated reproduction (Smith & Whitelam, 1997). Moreover, GAs help mediate shade-avoidance responses (Djakovic-Petrovic et al., 2007), in addition to activating floral organ identity genes (Yu et al., 2004) and influencing floral size (Goto & Pharis, 1999; Olszewski et al., 2002).

Based on these previous findings, we raised A. thaliana accessions under control (R : FR = 1.0–1.2, photosynthetically active radiation (PAR) = 40% ambient) and foliar-shade (R : FR = 0.6, PAR = 40% ambient) treatments in a split-plot experimental design (for discussion of light treatments, see Dechaine et al., 2009). On 19 January 2006, 3–5 seeds from each of 133 accessions (see Table S3 for list of accessions) were planted into 12 5.7 cm2 pots containing commercial potting mix (Sunshine #5; Sun Gro Horticulture, Bellevue, WA, USA) and were stratified in the dark at 4°C for 4 d to facilitate synchronous germination. Twelve pots were then randomly assigned to six plots of each light treatment (i.e. 2 treatments × 133 accessions × 6 replicates/accession/treatment) and thinned to one plant per pot following germination. Plants were censused daily for flowering, and, the following day, the most apical newly-opened flower with reflexed petals was collected (10:00–12:00 h) and preserved in 70% ethanol.

Flowers were dissected under a stereomicroscope (SMZ800; Nikon, Tokyo, Japan) and digitally photographed (Coolpix 5000; Nikon) once petals, stamens and the pistil were arranged flat on the stage. Using ImageJ software (ImageJ v.1.31; Wayne Rasband, National Institute of Health, Bethesda, MD, USA), we measured petal length, midpoint length (base of the petal to the point at which the petal blade reflexes), petal width, the filament and anther length of a long stamen, pistil length (from receptacle to stigma) and the style-stigma length (from the top of the ovary to the stigma). Finally, we used Sigma’s Extract-N-Amp Kit to extract DNA and amplify CAP/dCAP sequences (Table S2), assigning each accession to a genotypic state at each of the four GA1 SNPs.

Association mapping in accessions: statistical analyses

We used a mixed-model ANOVA (PROC MIXED, Cary, NC, USA SAS v.9.2) to partition variation in all measured floral morphological traits among the fixed factor, light treatment and the following random factors: subplot nested within treatment, accession and accession by treatment interaction. We detected significant genetic variation (accession term; see Table S4) but no effect of light treatment or the accession by treatment interaction. We estimated best linear unbiased predictors (BLUPs) of each floral trait for accessions across both light treatments. Because floral traits are commonly genetically correlated in A. thaliana (Juenger et al., 2000; Brock et al., 2009), we performed a PCA on the BLUPs of all measured floral traits (PROC PRINCOMP, SAS v.9.2; Table S5a) to reduce the dataset into a single variable with which to test for associations with GA1 while accounting for the lack of independence among floral traits.

We used a mixed-model approach to explore associations between variation in GA1 and floral morphology (Yu et al., 2006). This analysis tests the relationship between genotypic variation (two common haplotypes or individual SNPs of GA1) and phenotypic variation in floral traits (α = 0.05 for this and subsequent statistical tests), while statistically accounting for multiple levels of relatedness that can produce spurious associations between genotype and phenotype (i.e. false positives; Pritchard & Donnelly, 2001). We incorporated two measures of relatedness as covariates in the ANOVA model (TASSEL; Bradbury et al., 2007). First, we estimated a kinship matrix (K-matrix) as the proportion of SNPs shared between accessions (i.e. identity by state; TASSEL, Bradbury et al., 2007). For background genotypic information, we used 149 SNPs genotyped from the A. thaliana accessions in our panel (Platt et al., 2010).

Second, we estimated population structure covariates (Q-matrix) using Bayesian Inference (Structure v.2.2; Pritchard et al., 2000a), which estimates the proportional relatedness of each accession to inferred ancestral populations in the panel of A. thaliana using the 149 SNPs (above). In order to determine the appropriate number of ancestral populations (K), we ran the model 20 times for = 1–15 for 150 000 iterations, following a burn-in of 50 000 iterations. Because A. thaliana is largely a selfing species and our genotypic data indicated extensive homozygosity, we estimated genetic structure using the ‘haploid’ setting while assuming that ‘allele frequencies are correlated’ and that there is ‘admixture’ among populations (Nordborg et al., 2005). We utilized two approaches when selecting K; first, we selected K with the highest average log-likelihood score (here; = 10; Table S6), and second, we used the ad-hocΔK approach developed by Evanno et al. (2005; here, the ΔK method indicates K = 2). In order to examine how these two K-selection methods control the false positive rate, we estimated the cumulative distribution of P-values using the 149 background SNPs (independent variable) and the floral first principal component (floral PC1; dependent variable) in the following ANOVA models; naïve (no controls for genetic structure), only the Q-matrix (Q2 or Q10; i.e. = 2 and 10, respectively), only the kinship matrix (K), or both Q- and K-matrices (Q2 + K or Q10 + K).

These comparisons of statistical controls in our data indicate that the kinship matrix largely controls Type I errors (Fig. S2) and that the Q-matrix provides little additional statistical control of false positives. However, previous studies in A. thaliana have shown that Q + K ANOVA models (maximizing likelihood when selecting K) optimally control false positives (Zhao et al., 2007; Ehrenreich et al., 2009). Consistent with these previous methodologies, inclusion of structure covariates in our analyses (i.e. Q10 + K) accounts for significant variation in the first principal component of measured floral traits. We proceed with the mixed-model approach (Yu et al., 2006) accounting for K- and Q-matrices (= 10) when examining GA1-floral associations.

We used TASSEL to implement the mixed-model analysis, testing the common GA1 haplotypes (GA1A vs GA1B) for an association with the floral first principal component. For significant associations, we subsequently test for GA1 associations of the two common GA1 haplotypes with each measured floral trait to explore which floral trait(s) contributes to the GA1-PC1 association. Where haplotypes were significantly associated with floral traits, we ran each of the four SNPs in separate analyses to examine the strength of their associations with floral morphological traits. We should note that these SNPs are not independent from each other – a consequence of their close physical proximity, which is reflected in significant estimates of LD among these four SNPs (for all pairwise estimates P < 0.0001; r2 ranges from 0.24 to 0.87 for accessions in the association mapping panel; Table S7). Despite the lack of independence, we do, however, view these analyses as an heuristic tool to explore which SNP (or SNPs) might contribute to the observed haplotype associations.

Despite our incorporation of kinship and structure matrices to statistically control for spurious effects of population structure, we cannot discount the possibility that, in our association panel of 133 accessions, GA1 is in tight physical linkage with a neighbouring causal locus. In order to explore the frequency of other significant SNP-floral associations in the region, we selected SNPs that are within a 40 kb region centred on GA1. SNPs were identified by merging the A. thaliana 250 k dataset (Kim et al., 2007; Atwell et al., 2010) with accessions in our glasshouse study. Unfortunately, only 68 accessions (of the 82 that carry a common GA1 haplotype) were present in the SNP dataset. Across this 40 kb region, median LD between GA1 SNPs and those 10 kb up- and downstream (i.e. SNPs between 0 and 10 kb and 30–40 kb) fell to r2 = 0.16 and 0.10, respectively. To limit possible effects of sampling error, we retained SNPs with a minimum sample size of 15 for the rare SNP state (N = 77 in the 40 kb region) and our haplotype data at GA1. We test for genotype–floral PC1 associations using mixed-model methods described above; however, results should be interpreted with caution due to the reduced total sample size.

Segregation analysis in MAGIC lines: phenotypic measurements

Variation in the two most common GA1 haplotypes (and many of the underlying SNPs) was significantly associated with A. thaliana floral morphological traits (see Results section). To further account for possible spurious association stemming from cryptic population structure (Pritchard et al., 2000b) or LD between GA1 and a distant causal locus, we utilized the Multiparent Advanced Generation Inter-Cross (MAGIC) set of A. thaliana lines (Kover et al., 2009) to test for supporting evidence of GA1-floral morphological associations. The MAGIC lines (MLs) are the product of four generations of random matings among 19 A. thaliana accessions, producing 342 F4 families from which three MLs per family were advanced by single-seed descent for six generations. The MAGIC population has negligible levels of LD among loci on different chromosomes, and the numerous recombination events during intermating and subsequent selfing reduced LD to 0.17 for loci below 500 kb on average genome-wide (Kover et al., 2009).

To phenotype floral morphological traits in the MLs, we raised three replicate plants of each of 189 lines in a completely randomized design glasshouse experiment (University of Wyoming, Laramie, WY, USA). On 5 December 2008, seeds were planted in pots (6.5 × 6.5 × 9 cm) containing Sunshine #5 potting mix. Following stratification in the dark for 4 d (4°C), pots were placed in the glasshouse under supplemental lighting (13 h light : 11 h dark) at 21°C and seedlings were later thinned to one per pot. Plants were censused daily for flowering, and the following day, two newly-opened flowers with reflexed petals were collected (10:00–12:00 h) and preserved in 70% ethanol. We dissected, imaged and measured floral traits as described above. Lines were genotyped at the four GA1 SNPs using methods described previously. Due to repeated failed reactions, seven lines had incomplete genotypic data and were dropped from the analysis (final sample size; N = 182).

Segregation analysis in MAGIC lines: statistical analyses

We used ANOVA to test whether MLs, designated as a random factor, influenced the average expression of measured floral traits (PROC MIXED). We detected significant genetic variation for all traits (see the Results section) and estimated BLUPs for each line. We again utilized PCA (PROC PRINCOMP; Table S5b) to reduce the dataset into a single variable that explains a majority of the variation in floral morphology and then used ANOVA (PROC GLM; SAS v.9.2) to test whether this floral variation among MLs segregates with that of GA1. Due to significant floral PC1-GA1 results, we subsequently tested associations between GA1 and each individual floral trait. As in previous analyses, we partitioned variation between the two common haplotypes (GA1A and GA1B; N = 51 and 59, respectively) and where significance was detected we further tested floral associations with GA1 SNPs for the genotyped 182 ML dataset. Reflecting the lack of independence among these SNPs in GA1, we did not observe recombinants between GA1_2040 and GA1_2354 in this population, and, as a consequence, these two SNPs were combined into a single analysis.

A bivariate genetic correlation matrix (i.e. the standardized G-matrix) is one measure of the genetic architecture among functionally related traits. We estimated genetic correlations among all measured floral traits using floral trait BLUPs of each MAGIC line (PROC CORR; SAS v.9.2). In this segregating progeny, significant genetic correlations among traits will arise from the pleiotropic action of individual genes and/or LD between independent, but physically proximate, genes. Due to multiple generations of intercrossing, this correlation matrix should be representative of floral genetic architecture independent of the potential influence of population structure in natural populations.

Segregation analysis of a single SNP in Bay-0 × Sha RILs: phenotypic measurements

Association mapping and segregation analyses of individual SNPs suggested that GA1 (GA1_2354) may contribute to the observed association between GA1 and floral morphology (see the Results section), although without additional experimental support we cannot discount the possibility that GA1 is in LD with a neighbouring causal locus. Moreover, linkage across GA1 (see Table S7) confounded our ability to test the association between floral morphology and a possible causal SNP (or SNPs). To further explore the possible influence of GA1_2354 on variation in the length of floral traits, we raised recombinant inbred lines (RIL) from the Bay-0 × Sha cross (Loudet et al., 2002), which segregate only for this SNP in GA1. Bay-0 (Arabidopsis Biological Resource Center stock number, cs6608) carries the GA1A haplotype, while Sha (cs6180) carries a rare allele that is identical to GA1A at the first three SNPs but differs at GA1_2354 (T vs. A, respectively). A significant association between GA1_2354 variation and floral traits in this RIL population would support the hypothesis that this individual SNP contributes to the overall GA1–floral length association. Additionally, because Bay-0 and Sha have different histories of recombination compared with the 19 parental lines of the MAGIC population, the pattern of LD surrounding GA1 likely differs – providing independent support for an association between GA1 and floral morphological variation.

Although we did not detect accession × light treatment interactions for floral morphology (or corresponding GA1 × light treatment interactions; data not shown), GA1–morphology associations were indeed stronger (i.e. F-values were larger) for the 133 accessions raised in the foliar shade treatment when analyses were performed separately by light treatments. To further explore the possibility that the GA1 association with floral morphology might differ across light environments, we raised these RILs in both control and foliar shade environments in growth chambers.

We raised one replicate plant of 49 Bay-0 × Sha RILs in each of two replicate growth chamber compartments per light treatment (N = 196). Seeds of each RIL were randomly planted into cells of araflats (Betatech bvba; Gent, Belgium) containing Sunshine #5 potting mix and were placed in the dark for 4 d (4°C). Following stratification, flats were randomized across four growth chamber compartments (Percival PGC-9/2; Geneva Scientific, Fontana, WI, USA) set to 20°C with control lighting conditions (R : FR = 1.2; PAR = 155 μmol m−2 s−1; 13 h light : 11 h dark) produced from fluorescent (Phillips F32T8/TL741) and incandescent (GE 15 W) bulbs. After 7 d the majority of RILs had germinated, and in two compartments, the foliar-shade treatment was imposed (R : FR = 0.6; PAR = 155 μmol m−2 s−1; 13 h light : 11 h dark) by adding wavelengths from FR incandescent bulbs (F32T8/IR-750; Geneva Scientific). Plants were monitored for first-flowering date, after which a newly opened flower with reflexed petals was harvested (4–5 h after subjective dawn) and preserved in 70% ethanol. Later, flowers were dissected, imaged and measured as described previously.

Segregation analysis of a single SNP in Bay-0 × Sha RILs: statistical analyses 

We used a two-way ANOVA to test whether the fixed factor, the light treatment or the random factors – compartment nested within light treatment, RIL, or the light treatment by RIL interaction – influenced measured floral traits (PROC MIXED). We detected significant genetic variation for all floral traits among the RILs but no effect of light treatment or the interaction between light treatment and RIL (see the Results section), and, as a result, we estimated RIL BLUPs for each floral trait across both light environments. To examine whether the SNP GA1_2354 associates with variation in floral traits in this segregating progeny where the remaining GA1 SNP sites are monomorphic, we partitioned variation in the first floral principal component (PROC PRINCOMP; Table S5c) among SNP states in GA1_2354 (PROC GLM). We further tested each individual floral trait to examine which trait might underlie the significant floral PC1–GA1 association.

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Quantitative genetics of floral morphology

We detected significant genetic variation for all measured floral traits in the panel of A. thaliana accessions (P < 0.0001 for all floral traits; Table S4), in the MAGIC lines (P < 0.0001 for all floral traits; Supplementary Material Table S8), and in the Bay-0 × Sha RILs (P < 0.004 for all floral traits; Supplementary Material Table S9). On average, broad-sense heritabilities of floral traits were high for the MAGIC lines (mean = 0.63; range 0.52–0.74) and were moderate for the accessions of A. thaliana (mean = 0.31; range 0.18–0.47) and the Bay-0 × Sha RILs (mean = 0.35; range 0.19–0.54).

We did not detect any influence of the light treatment or the line accession × light treatment interaction on floral morphological traits in the A. thaliana accession panel (P > 0.10 for light treatment on all floral traits; P > 0.18 for the line accession × treatment interaction on all floral traits) or in the Bay-0 × Sha RILs (P > 0.66 for light treatment on all floral traits; P-values not estimable for the line × treatment interaction on all floral traits). The lack of a detectable effect of light environment (or line × treatment interaction) on these floral traits may be a consequence of the reduced power of the split-plot design relative to other experiments where effects of light quality (R : FR) have been detected (Brock & Weinig, 2007).

Bivariate genetic correlations among floral traits within the MAGIC lines were strongly positive and significantly different from zero for all pairwise comparisons, except between anther and style-stigma length (Table 2). These correlations result from pleiotropy or from LD between loci in close physical proximity. This integrated genetic architecture (i.e. occurrence of strong pairwise correlations for all traits) is supported by previous studies of A. thaliana (e.g. Juenger et al., 2000; Brock & Weinig, 2007) and other Brassicaceous species (e.g. Conner & Via, 1993; Brock et al., 2010).

Table 2.   Bivariate genetic correlations between all measured floral traits (LN, length; WD, width) in a segregating progeny resulting from Multi-parent Advanced Generation Intercrosses (MAGIC) between 19 Arabidopsis thaliana accessions
 Midpoint LNPetal WDFilament LNAnther LNPistil LNStyle-stigma LN
  1. Significant difference: ***, P < 0.001; ****, P < 0.0001.

Petal LN0.87****0.66****0.71****0.41****0.70****0.48****
Midpoint LN0.41****0.72****0.26****0.70****0.39****
Petal WD 0.28****0.40****0.41****0.30****
Filament LN  0.23***0.68****0.38****
Anther LN   0.26***0.12
Pistil LN    0.65****
Style-stigma LN     

GA1 molecular variation

In 32 accessions of A. thaliana, we observed 24 polymorphic sites across the GA1 coding region of which 14 were nonsynonymous polymorphisms (of which five were singletons). We focused our association analyses on four nonsynonymous polymorphisms (Table 1a), where the minimum frequency of rare SNP states was equal to (or greater than) 0.25 in our initial survey of sequence data. These four SNPs were genotyped in 133 A. thaliana accessions and segregated at intermediate frequencies (mean minimum frequency = 0.4; range (0.32–0.5; Table 1a). Additionally, the two most common haplotypes of these SNPs (GA1A and GA1B) had moderate replication (Table 1b), and phylogenetic analyses illustrate that they occupied distinctly divergent basal branches (Fig. S1b). Based on the frequency and abundance in our A. thaliana panel, individual SNPs and the two common haplotypes had sufficient replication to test for associations with floral morphological traits.

GA1–floral associations

In the panel of A. thaliana accessions GA1 was significantly associated with variation in the first principal component of floral morphological traits (floral PC1) and in the length of several floral traits (Table 3, Fig. 1), even after statistical controls for population structure and kinship. The two common GA1 haplotypes were strongly associated with petal length and midpoint length (the base of the petal to the point at which the petal blade reflexes), explaining 4.2% and 8.2% of trait variance, respectively. These haplotypes were also moderately associated with anther and style-stigma length variation (3.2 and 1.8 percent variance explained (PVE), respectively). Tests of association between floral traits and each SNP support results observed with the common GA1 haplotypes (Table 3). In general, SNP associations with petal morphology explained less variation than those observed with the haplotype analysis, although this pattern may be a consequence of sampling error.

Table 3.   Association analyses between GA1 and floral morphological traits, including the first floral principal component, and length (LN) or width (WD) of each floral trait, in a large panel of Arabidopsis thaliana accessions raised in the glasshouse
TraitGA1 haplotypesGA1_467GA1_1169GA1_2040GA1_2354
FPPVE (%)FPPVE (%)FPPVE (%)FPPVE (%)FPPVE (%)
  1. Variables consist of four individual GA1 SNPs (e.g., GA1_467; N = 133) and the two most common haplotypes of these SNPs (N = 86). Additional details on SNPs is provided in the Materials and Methods section (Table 1). For significant associations, we provide percentage variance explained (PVE) by the haplotype (or SNP).

Floral PC16.50.01264.03.70.05582.14.50.03512.44.90.02882.34.50.03592.3
Petal LN9.30.00324.23.70.0581.83.50.06291.54.20.04192.33.60.06132.0
Midpoint LN12.90.00068.26.40.01284.23.60.05922.47.20.00843.77.50.00733.8
Petal WD1.20.27
Filament LN1.60.21
Anther LN4.40.03933.21.90.172.30.144.80.032.04.10.04551.9
Pistil LN1.80.18
Style-stigma LN3.90.05121.80.010.914.80.02992.46.60.01151.75.80.01721.7
image

Figure 1. Boxplots illustrating the association of GA1 haplotypes (GA1A and GA1B) or individual GA1 SNPs (2354A vs. 2354B) with petal, midpoint, filament and anther length traits in an association panel of 133 Arabidopsis thaliana accessions, a segregating progeny produced by intercrossing 19 A. thaliana accessions (MAGIC lines), and the A. thaliana recombinant inbred line (RIL) set between Bay-0 and Sha accessions. In the association panel, we illustrate haplotype differences with floral trait residuals produced from a model that accounted for population structure and kinship among A. thaliana accessions. Closed boxes indicate significant associations; open boxes indicate marginally significant associations (P-value < 0.07). LN, length.

Download figure to PowerPoint

Cryptic population structure or linkage disequilibrium between GA1 and a potentially causal locus could result in spurious GA1–floral associations, despite our use of statistical covariates to control for multiple levels of relatedness. To corroborate our association results, we tested for co-segregation between floral traits and variation in GA1 in the Arabidopsis MAGIC population: a segregating progeny with minimal LD. In support of our association analyses, we observe that floral PC1, petal, midpoint, filament and anther length co-segregate with the two common haplotypes of GA1 (Table 4, Fig. 1). However, we did not find evidence that GA1 associates with style-stigma length. Interestingly, there was variation in the identity of GA1 SNPs that segregated with floral morphological traits. More specifically, petal and stamen traits segregated most strongly with GA1_2040 and GA1_2354 (Fig. 1), but not at all with GA1_467. A similar, albeit weaker, pattern was observed in the number and significance of individual GA1 SNP associations in the panel of A. thaliana accessions (Table 3). Taken together, these results suggest that SNPs at the 3’ end of the coding sequence may play an important functional role in the observed GA1–floral associations.

Table 4.   Segregation analyses between genetic variation in GA1 (haplotypes and individual SNPs; N = 110 and 182, respectively) and floral morphological traits, including the first floral principal component, and length (LN) or width (WD) of each floral trait, in the Arabidopsis Multi-parent Advanced Generation Intercross (MAGIC) lines
TraitGA1 haplotypesGA1_467GA1_1169GA1_2040 & GA1_2354
FPPVE (%)FPPVEFPPVE (%)FPPVE (%)
  1. Percentage variance explained (PVE) provided for significant results.

Floral PC15.70.01925.11.30.265.90.01633.27.60.00644.1
Petal LN4.80.03124.21.70.194.80.03022.65.90.01613.2
Midpoint LN3.70.05763.31.00.333.00.08481.74.20.04192.3
Petal WD0.010.92
Filament LN6.50.0125.71.60.213.80.0522.17.80.00594.1
Anther LN4.60.03374.11.70.196.60.01093.59.50.00235.0
Pistil LN1.30.26
Style-stigma LN1.50.23

Using the Bay-0 × Sha RILs, we were able to test for co-segregation between floral morphology and the GA1_2354 SNP in a segregating progeny that is monomorphic for the remaining three GA1 SNPs. We detected co-segregation between the GA1_2354 SNP and floral PC1, petal, midpoint, filament and pistil length (Table 5, Fig. 1). In addition to supporting the general GA1-floral morphological association, these results suggest that GA1_2354 may be a causal SNP underlying at least part of the observed association. Moreover, examination of GA1–floral associations across these three populations appears to reveal that while aspects of petal and stamen morphology are consistently associated with GA1, its associations with pistil traits may differ subtly with environmental or experimental error.

Table 5.   ANOVA results of variation in GA1_2354 and all floral traits, including the first floral principal component, and length (LN) or width (WD) of each floral trait, in the Bay-0 × Sha segregating progeny (N = 49)
GA1_2354FPPVE (%)
  1. Parental accessions are fixed with the GA1A haplotype for the remaining SNPs and only segregate for this final SNP. Percent variance explained (PVE) provided for significant associations.

Floral PC110.30.002417.9
Petal LN8.70.004915.6
Midpoint LN18.8< 0.000128.6
Petal WD0.20.68
Filament LN3.50.0688 6.9
Anther LN0.10.78
Pistil LN9.30.003816.5
Style-stigma LN1.30.26

SNP–floral associations in the GA1 region

We merged our GA1 common haplotype data with publically available SNP datasets in order to explore the extent of SNP–floral associations in the vicinity of GA1. Of the 78 SNPs tested in the 40 kb region centred on GA1 (Supplementary material Table S10), 66 showed no association with floral morphology. Twelve showed significant associations with flower shape, and a majority of the significant SNPs (N = 7) reside within GA1 or in adjacent upstream and downstream intergenic regions. The remaining five ‘non-GA1’ SNPs that were also significantly associated with floral morphology could be causal and might account for a spurious GA1–floral association. However, publically available re-sequencing data coupled with floral data in RILs indicate that three of the five SNPs are unlikely to be causal, because they are either polymorphic in RILs that lack floral QTLs in the GA1 region or, conversely, they are monomorphic in RILs with floral QTLs in the GA1 region (Table S10). The remaining two SNPs associating with floral PC1 are in noncoding regions, but it is nevertheless possible that one of these SNPs (or other untested SNPs in LD with GA1) could underlie the observed GA1–floral association.

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Floral morphology plays a central role in plant fitness, primarily through effects on the efficiency of pollen movement; correspondingly, a large number of studies have investigated the quantitative genetics of floral morphology with the aim of shedding light on the evolutionary potential of these traits (Conner & Via, 1993; Galen, 1996; Ashman, 2003; Gomez et al., 2009). In parallel to these evolutionary studies, molecular-genetic studies using knockout and expression mutants have identified many of the genes that influence the initiation and development of floral organs (Coen & Meyerowitz, 1991; Luo et al., 1996; Yu et al., 2004; Szecsi et al., 2006). These latter mechanistic studies provide an important list of candidate loci that may underlie quantitative variation in floral traits in natural populations. However, despite these evolutionary and developmental advances, few loci contributing to natural variation in floral organ morphology have been identified (but see Cubas et al., 1999), and thus the genes that may contribute to standing variation in natural populations and potentially to adaptive floral evolution remain unknown. A range of approaches has been pioneered to begin linking the fields of evolutionary and developmental genetics, and to begin identifying the QTL and QTN and ultimately networks that affect quantitative-trait expression (reviewed in Mackay et al., 2009). Using a candidate-gene approach, we identify strong associations between floral organ length and GA1 and confirm these associations with multiple segregating progeny.

The locus GA1, which catalyses the first step in the production of gibberellins, emerged as a candidate for controlling floral organ size as a consequence of several lines of evidence. As noted above, previous molecular genetic studies demonstrate that null mutations in GA1 pleiotropically affect the size of floral organs, especially in petals and stamens (Koornneef & Vanderveen, 1980; Goto & Pharis, 1999; Yu et al., 2004; Hu et al., 2008) – a phenotype rescued by addition of bioactive GA. In the Arabidopsis flower, GA is synthesized either in the stamen or receptacle (Cheng et al., 2004; Hu et al., 2008), which co-localizes with the expression pattern of GA1 (Silverstone et al., 1997); once synthesized, GA is transported to petals for proper growth. Similar results have been found in other systems suggesting that GA is widely important to floral development. For example, anther removal in Petunia hybrida retards corolla growth, which can also be rescued by the addition of GA (Weiss & Halevy, 1989). Finally, recent studies in A. thaliana indicate that during late-stage floral development, bioactive gibberellins promote the expression of B- and C-class genes (AP3, PI, and AG; of the ABC model), which, in turn, regulate petal, stamen and pistil development (Coen & Meyerowitz, 1991; Yu et al., 2004).

Our findings extend these results, and suggest that GA1 accounts for naturally occurring quantitative variation in the size of floral organs. Our analyses of natural variation at this locus reveal two well-defined groups of GA1 variants (Fig. S1b) characterized by four SNPs. We observe that variation in GA1 is strongly associated with variation in petal and stamen length and to some extent pistil length in an association panel of 133 accessions of Athaliana. This association is further supported in the MAGIC population derived from interbreeding of 19 parents and in a separate set of segregating progeny developed from the Bayreuth × Shahdara accessions. The observed co-segregation in these two experimental populations strongly suggests that GA1 is the causal locus because the structure of LD in the two alternative progeny sets is likely to differ, that is, an unanalysed causal locus proximate to GA1 is unlikely to have confounding patterns of LD in both segregating progenies. If we assume that the causal locus is GA1, the co-segregation results from the Bay-0 × Sha RILs implicate the 4th SNP (GA1_2354) as a QTN, because, based on resequencing data from Bay-0 and Sha (http://www.1001genomes.org), this is the only SNP in GA1 (between 5.6 kb upstream of the start codon to the stop codon) that segregate in this progeny set. In support of a causal effect of GA1_2354, substitution matrices suggest that amino acid replacements at this site may be disfavoured (Henikoff & Henikoff, 1992); however, the weaker GA1_2354-floral associations relative to those observed in the haplotype analysis (Table 3) suggest that the putative effects of GA1 on floral morphology likely arise from a combination of GA1_2354 and other SNPs (e.g. GA1_1169; a predicted disfavoured nonsynonymous polymorphism near the GA1 active site; Koeksal et al., 2011). Intriguingly, Juenger et al. (2000) may have identified GA1 as a petal length QTL in a study where QTL for floral organ size was mapped in the Ler × Col mapping population (Lister and Dean 1993), which are GA1B and GA1A, respectively, and the additive effect of the QTL alleles were in the same direction as observed here (Fig. 1). This QTL was not mapped in the Ler × Cvi mapping population (Alonso-Blanco et al., 1998; Juenger et al., 2005), which is invariant for the 4th SNP in GA1.

The percentage variance explained by the common GA1 haplotype associations in the A. thaliana accession panel and in the MAGIC lines average 4.3% (range 1.8–8.2%), while the final SNP, GA1_2354, averages 3.5% (range 2.3–5.0%). The percentage variance explained by GA1 is thus small to moderate; however, with sample sizes < 500, effect sizes can be overestimated because weak effects below the critical threshold are not included, leading to a truncated distribution of effect sizes (Beavis, 1998). It is noteworthy that this effect size is comparable to what many studies are revealing for the genetic architecture of complex traits (Buckler et al., 2009; Mackay et al., 2009). Furthermore, the percentage variance explained is comparable to that explained by other floral organ length QTL (Juenger et al., 2000, 2005; Hall et al., 2006), including the QTL mapped in Athaliana in the genomic region of GA1 (Juenger et al., 2000). These percentage variances explained are consistent with basic quantitative genetic theory (Fisher, 1930; Falconer & Mackay, 1996), which would predict that adaptive evolution in floral traits occurs via allele frequency changes at comparatively small effect size loci.

Despite the extensive support of a GA1–floral morphology association, without additional transgenic complementation experiments or fine-scale mapping we cannot discount the possibility that GA1 is in LD with a physically proximate causal locus. Because a subsample of the accessions carrying the common GA1 haplotype have been SNP genotyped (Kim et al., 2007; Atwell et al., 2010), we were able to explore the extent of significant SNP–floral PC1 associations in the region surrounding GA1. The vast majority of SNPs surrounding but outside GA1 (66 of 78) showed no association with floral morphology. Of those associations that were significant, a majority (7 of 12) are within or adjacent to the GA1 coding region. We also detected five other significant associations; however, discordant results between the presence/absence of sequence variation and floral QTL suggests that the three SNPs in the coding regions of genes neighbouring GA1 are not causal (Bay-0 × Sha RILs, data presented here; Ler × Cvi, Juenger et al., 2005) and the two remaining SNPs were in noncoding regions (although these could have an effect via splicing or gene regulation). While some SNPs in the region are likely to be in LD with GA1 by random chance, we cannot discount the possibility that these two noncoding SNPs (or other untested SNPs) underlie the detected GA1–floral association. Despite these caveats, we would again note that the majority of significant SNP–floral associations in this 40 kb region are centred on GA1.

Variation in GA1 was present not only at the large geographic scale sampled for the association panel, but also within segregating progeny populations. In our association panel, in instances where two (or more) accessions were genotyped from the same population (e.g. Bu-0 and Bu-14 are GA1A and GA1B, respectively), variation in at least one of the four GA1 SNPs was present in slightly > 50% of the populations. This observation of segregating variation within populations of Athaliana in combination with the apparent association between this locus and floral-organ size suggests that evolution of floral morphology could occur in natural populations of A. thaliana via shifts in allele frequency at GA1. For example, in selfing populations of A. thaliana, smaller-flowered plants (i.e. the putative GA1A phenotype) might be advantageous due to the reduction in costs associated with pollinator attraction (Charlesworth & Charlesworth, 1987; Lloyd, 1987). Also, preliminarily, it is worth noting that GA1 might affect floral organ sizes in another related species, Brapa. A QTL for midpoint length maps to chromosome 3 (Brock et al., 2010) and BLAST searches indicate that one of two putative GA1 copies co-localizes to the midpoint 2-LOD interval (P. Lou, pers. comm.).

With regard to a general model for quantitative variation in floral organ size and allometry, it may be that loci will commonly affect organs in more than 1 whorl, consistent both with the extensive pleiotropy of ABC genes regulating qualitative variation in floral organ identity (Coen & Meyerowitz, 1991) and with broad effects of most hormones. The current results are coincidentally consistent with the ABC model as well as previous studies (Koornneef & Vanderveen, 1980; Goto & Pharis, 1999; Cheng et al., 2004) in that GA1 affects organs in the adjacent 2nd and 3rd whorls more strongly than those in the 4th whorl. However, the nearly identical genetic correlations observed between petal, filament and pistil lengths (range: 0.68–0.71; Table 2) indicates that other loci must counterbalance the effect of genes such as GA1 and contribute to the similar magnitude of genetic correlations typically observed among all floral organs (Juenger et al., 2000, 2005; Brock & Weinig, 2007). The differential degree to which GA1 associates with mean midpoint length (i.e. the petal region where the blade reflexes) and mean filament length (Fig. 1) also suggests that this locus contributes not only to variation in individual floral organs, but also to floral allometry (relative scaling between floral organs) as well. These effects on allometry have evolutionary implications. In related outcrossing species, the distance that anthers extend from the opening of the corolla (stamen exsertion) has been shown to influence pollen export and thus male fitness (Morgan & Conner, 2001; Conner et al., 2009). Interestingly, recent evidence suggests that A. thaliana populations may experience higher outcrossing rates than previously realized (up to 14.5% effective outcrossing; Bomblies et al., 2010). In outcrossing species (and possibly in A. thaliana), GA1 and other loci (e.g. DELLA genes that repress GA responses, Cheng et al., 2004; Yu et al., 2004) that also influence B- and C-class genes in the ABC floral model may partially underlie expression of floral traits involved with pollen export and receipt, all of which are points that require further investigation.

Acknowledgements

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

The authors appreciate the help of J. Dechaine, L. Demink, L. Lucas, A. McKnite, B. Wehmeyer and C. Willis in phenotyping and genotyping Arabidopsis plants; Z. Gompert and A. Buerkle for assistance in scripting the estimation of STRUCTURE covariates. This research was supported by NSF grant DBI-0227103 to C.W. and by a BBSRC grant to P.X.K.

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Supporting Information

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Fig. S1 A graphical representation of the GA1 locus of Arabidopsis thaliana and a Bayesian phylogram of full-length GA1 sequences including both exon and intron regions from 32 accessions of A. thaliana rooted by A.  lyrata.

Fig. S2 Cumulative distribution of P-values using the 149 background SNPs to test for associations with the floral first principal component.

Table S1 List of 32 Arabidopsis thaliana accessions, location of origin (including latitude and longitude), and GenBank number for the GA1 sequence utilized in exploring levels polymorphisms and estimating GA1 genealogical trees

Table S2 Primer sequences for four SNPs identified in GA1 and enzymes for associated restriction digests

Table S3 List of 133 Arabidopsis thaliana accessions used in candidate gene association analyses between floral morphology and GA1

Table S4 Mixed-model ANOVA of floral morphological traits measured from 133 Arabidopsis thaliana accessions

Table S5 Principal component analysis of all measured floral traits in (a) an association mapping panel of Arabidopsis thaliana, (b) the Arabidopsis MAGIC segregating progeny, and (c) the Bay-0 × Sha recombinant inbred lines

Table S6 Average likelihood values and standard deviation from 20 runs of STRUCTURE (K = 1–15) examining population structure in 133 Arabidopsis thaliana accessions genotyped at 149 SNPs

Table S7 Linkage disequilibrium among four SNPs genotyped at the GA1 locus in 133 accessions of Arabidopsis thaliana

Table S8 ANOVA of floral morphological traits measured from the Arabidopsis MAGIC segregating progeny

Table S9 Mixed-model ANOVA of floral morphological traits measured from the Arabidopsis thaliana Bay-0 × Sha recombinant inbred lines

Table S10 Association analyses between 77 SNPs (plus the GA1 haplotype described in the Materials and Methods section) within the 40 kb region centered on GA1 and the first floral principal component

Please note: Wiley-Blackwell are not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing material) should be directed to the New Phytologist Central Office.

FilenameFormatSizeDescription
NPH_4145_sm_FigS1-S2.docx2144KSupporting info item
NPH_4145_sm_TableS1-S10.docx61KSupporting info item