Arabidopsis semidwarfs evolved from independent mutations in GA20ox1, ortholog to green revolution dwarf alleles in rice and barley

Significance

Semidwarf accessions occur at low frequency across the distribution range of Arabidopsis thaliana and are mainly mutants of the GA5 (GA20ox1) gene, mutations of which originate from wild-type alleles still present in the regions where the mutants were found. We identified the causal mutations by allelism tests and sequencing and performed a detailed population genetics analysis of this variation. Using Fay and Wu’s H statistics, we obtained indications for local selection of the dwarf alleles. Mutants of functional orthologs of this gene have been selected as the so-called “green revolution genes” in rice and barley, thus indicating that Arabidopsis natural variation can be a source for the identification of useful genes for plant breeding.

Abstract

Understanding the genetic bases of natural variation for developmental and stress-related traits is a major goal of current plant biology. Variation in plant hormone levels and signaling might underlie such phenotypic variation occurring even within the same species. Here we report the genetic and molecular basis of semidwarf individuals found in natural Arabidopsis thaliana populations. Allelism tests demonstrate that independent loss-of-function mutations at GA locus 5 (GA5), which encodes gibberellin 20-oxidase 1 (GA20ox1) involved in the last steps of gibberellin biosynthesis, are found in different populations from southern, western, and northern Europe; central Asia; and Japan. Sequencing of GA5 identified 21 different loss-of-function alleles causing semidwarfness without any obvious general tradeoff affecting plant performance traits. GA5 shows signatures of purifying selection, whereas GA5 loss-of-function alleles can also exhibit patterns of positive selection in specific populations as shown by Fay and Wu’s H statistics. These results suggest that antagonistic pleiotropy might underlie the occurrence of GA5 loss-of-function mutations in nature. Furthermore, because GA5 is the ortholog of rice SD1 and barley Sdw1/Denso green revolution genes, this study illustrates the occurrence of conserved adaptive evolution between wild A.thaliana and domesticated plants.

Bioactive gibberellins (GAs) are plant growth regulators involved in important traits such as seed germination, flowering time, flower development, and elongation growth (1). GA biosynthesis and signaling pathways are well defined (1, 2) and have been targeted in crop breeding. Modification of GA pathways was crucial in the green revolution because it conferred semidwarfness, thus reducing lodging and increasing crop yields (3–6). Green revolution semidwarf varieties in wheat are due to mutations in DELLA genes, whereas many short straw rice varieties carry a mutation in the Semi-Dwarf-1 (SD1) locus. This locus codes for GA 20-oxidase-2, a GA biosynthesis gene that is also mutated in most modern barley varieties in which the gene was called Denso or Semi-dwarf 1 (Sdw1) (7).

GA 20-oxidases are involved in the later steps of GA biosynthesis and belong to the group of 2-oxoglutarate–dependent dioxygenases that, together with GA 3-oxidases, form biologically active GA (8). Arabidopsis thaliana has five GA20ox paralogous genes. AtGA20ox-1, AtGA20ox-2, AtGA20ox-3, and AtGA20ox-4 can catalyze the in vitro conversion of GA₁₂ to GA₉. Therefore, GA20ox paralogs might have partial redundant functions (9). However, among paralog genes, only AtGA20ox-1 (GA5), which was cloned on the basis of the ga5 mutant (10), affected plant height (8).

Natural variation for GA biosynthesis has been previously described in A. thaliana because the Bur-0 accession carries a loss-of-function allele at GA20ox4 (9), which does not result in a semidwarf phenotype. In addition, genetic variation in GA1 has been associated with variation in floral morphology (11). Furthermore, the semidwarf phenotype (here defined as a plant height shorter than half the size of genetically related individuals) observed in the Kas-2 accession is due to a recessive allele at the GA5 locus (12). The latter finding led to the questions of whether green revolution alleles, artificially selected in cereals, could also occur in natural populations of the wild species A. thaliana, and if so, how many different GA5 loss-of-function alleles exist, how they are distributed, and why they occur in some populations.

Results

Identification, Characterization, and Geographic Distribution of Natural ga5 Alleles.

Phenotypic surveys for plant height in worldwide collections of A. thaliana accessions detected 97 individuals collected in 23 different locations showing semidwarf phenotypes. To determine the genetic basis of semidwarfness, we carried out allelism tests by crossing at least one semidwarf from each population to the recessive ga5 (Ler) mutant (13) and to Ler “wild type” as control (Fig. 1 A and B and SI Appendix, Table S1). To discard the possibility that GA biosynthesis mutations other than GA5 could account for the semidwarf phenotypes, we tested the complementation of the ga5 (Ler) mutant by two other semidwarf mutant alleles also affecting GA biosynthesis: ga4 (Ler), a mutant in the GA3ox1 gene, and ga3ox1-3 (Col-0) (14) (Fig. 1B and SI Appendix, Table S2). Ler and Col mutants were used to test background effects. Control F₁ plants derived from crosses between nondwarf accessions and ga5 mutants and F₁ plants grown from crosses with other GA mutants were all taller than their corresponding parents. The crosses ga5 × ga4 and Ler × ga4 yielded a low height due to the erecta mutation, which remained recessive in the F₁. In addition, three accessions showing a weaker semidwarf phenotype (Nfro, Norway; Kar, central Asia; and Vel, Spain) were not allelic to ga5, which indicated that other loci accounted for their plant height phenotype. However, for all of the remaining semidwarf accessions tested, the F₁ obtained from the cross to ga5 exhibited the small size of the parents, whereas semidwarfism was lost in the cross with Ler. This finding confirmed the recessiveness of the semidwarf alleles. Therefore, most semidwarf accessions were allelic to ga5.

Fig. 1.

Semidwarf genotypes allelic to ga5 are present in nature. (A) Allelism test between the semidwarf mutant ga5 (13) and the semidwarf central Asian accession Dja-1. Neo-3 (central Asia) shows the phenotype of a functional GA5. Pictures were taken 2 wk after flowering. The upper left image shows the phenotype of ga5 and Ler at harvesting time. (Scale bars, 7 cm.) (B) Mean values of stem height ± SE in F₁ plants derived from crosses between ga5 or Ler and twenty accessions (Acc) allelic to ga5 (red), three nondwarf accessions (Col-0, Pak-1, and Cat-13), two semidwarf mutants (ga4 and ga3ox1-3), and one semidwarf accession nonallelic to ga5 (Vel-2). (C) Geographical distribution of semidwarf accessions in Europe, Scandinavia, and central Asia. Red marks indicate the locations of populations containing semidwarf accessions allelic to ga5. The right map shows the detailed local distribution of semidwarf (red) and wild-type (blue) individuals found in the Oude Wetering (OW) population from The Netherlands in 2012.

To evaluate if there is any general negative pleiotropic effect on plant performance associated with natural ga5 alleles, we measured several presumably adaptive traits in six wild ga5 semidwarf accessions and in the ga5 mutants in Ler and Col genetic backgrounds (SI Appendix, Fig. S1). Consistent with previous studies (8), ga5 mutants did not differ significantly from their wild types in the evaluated traits (SI Appendix, Fig. S1). However, natural ga5 accessions strongly differed in flowering time, branch number, and silique number, indicating the absence of strong ga5 effects on these traits but substantial contributions from other genes. Therefore, no major tradeoff on silique number, assumed to be a proxy for fitness, was found for these naturally occurring ga5 alleles.

Semidwarf ga5 accessions were found in 23 different populations distributed in western Europe, the Iberian Peninsula, Scandinavia, central Asia, and Japan (Fig. 1C and SI Appendix, Table S1). From our analysis, we roughly estimated that at worldwide scale, the frequency of wild populations containing semidwarf accessions allelic to ga5 was at least 1%. However, these frequencies may be higher: because most populations segregate for GA5 loss-of-function alleles, we cannot discard the possibility that some populations with a limited number of individuals may contain semidwarfs at low frequency not represented in the individuals studied. We also found a semidwarf frequency of 1% in the Hapmap experimental population consisting of 360 worldwide accessions with empirically reduced population structure (15). However, the frequency of ga5 semidwarf-containing populations was not homogeneous throughout the A. thaliana geographic range because we did not find semidwarfs among the many central and east European accessions studied. By contrast, semidwarfism appeared more frequently in central Asia than elsewhere; 5 out of the 24 central Asian populations monitored in this and another study (16) carried semidwarf individuals (SI Appendix, Table S1). A ∼2% frequency was estimated for the Iberian Peninsula from the qualitative analysis of the intensive collection (17) used to select the Iberian accessions included in this study. In addition, detailed sampling and analysis of ga5 semidwarfs in The Netherlands indicated a ∼5% frequency in this region. Dutch semidwarfs seemed to have spread mainly in the west of the country, although one population was found inland (Fig. 1C).

Descriptions of the habitat of populations containing ga5 semidwarf individuals show that they occur in multiple diverse environments where the species occurs. For instance, Dutch semidwarf accessions were found in the anthropoid environments where A. thaliana grows, including urban (street populations) and rural (road and field sides; SI Appendix, Fig. S2) habitats. However, in the Iberian Peninsula and central Asia, semidwarfs occurred in more natural environments, including Mediterranean forests and mountain wet grasslands (SI Appendix, Fig. S2). This wide geographic and ecological distribution indicates that ga5 semidwarfism does not show a strong geographic structure and is not associated with a single and common climatic factor across its distribution range.

Identification of Multiple GA5 Loss-of-Function Alleles.

To determine the putative mutations causing semidwarf phenotypes, we sequenced the GA5 gene (∼1.5 kb) in 59 semidwarf accessions collected worldwide and 135 nondwarf individuals, which were collected from the same population or geographic region as the semidwarfs identified. For the Dutch OW and Sch populations, the ∼1-kb GA5 region, spanning semidwarf causal mutations, was sequenced in 16 semidwarfs and 77 wild-type individuals. Collectively, sequencing data identified 21 different mutations which were predicted to cause GA5 loss-of-function alleles in semidwarf accessions (Fig. 2). These mutations were classified in six loss-of-function classes according to their nucleotidic nature. First, nonsense mutations causing premature stop codons were found in Kas-2 and Sparta. Second, missense mutations were found close to the conserved metal binding sites of GA5 in an Iberian (Mar-1, Mar-3, and Mar-11) and a Scandinavian (Var 2–1 and Var 2–6) population, which might underlie their ga5 phenotype. In addition, the Mdc-10 and Mdc-53 semidwarf accessions also carried missense mutations in GA5 conserved domains. Third, a single substitution in the donor splice site of the first intron was found in all Dutch semidwarf accessions. This affects normal GA5 splicing generating a truncated GA5 protein. Fourth, seven small insertions (Cat-0, Dja-1, and Pak-3) or deletions (Cat-17, Cat-23, Cat-43, and Sus-1) were predicted to cause frame shifts and truncated GA5 proteins. Fifth, a transposon insertion, with high similarity to At4g04410, was identified in the MdcA-60 accession. Sixth, several large deletions (>20 bp) were found in some accessions. These included a 29-bp deletion in the first exon of Kl-2 (Germany) and a 444-bp deletion spanning part of the second exon and the complete third exon of accession T1080 (Sweden) (SI Appendix, Figs. S3 and S4 and Table S3). This deletion was first detected by the absence of sequence coverage in the 1001 Genomes Project data (www.1001genomes.org) and further confirmed by extensive PCR amplifications (SI Appendix, Figs. S3 and S4 and Table S3). In addition, large GA5 deletions of several kilobases were found in the Veg 1–1, Kyr-2, and YGU accessions. These deletions included not only the coding region but also the promoter (SI Appendix, Figs. S3 and S4 and Table S3) and were associated with absence of GA5 expression in Kyr-2, Veg 1–2, and YGU.

Fig. 2.

Natural loss-of-function mutations in the AtGA20ox1 (GA5) gene. The different natures and positions of mutations causing GA5 loss-of-function alleles are shown in each row. Exons (black boxes) are connected with horizontal lines representing intronic regions of GA5. Iron and 2-oxoglutarate binding sites (18) are indicated at the top.

Sequencing analyses indicated that most populations containing semidwarf individuals carry a single loss-of-function mutation in all dwarf plants (e.g., OW-0 in Fig. 2). However, two Iberian populations (Cat and Mdc) segregated for four independent GA5 loss-of-function mutations (Fig. 2). One allele appeared more frequently; it was present in eight Cat individuals out of 22 sequenced samples. On the other hand, most GA5 loss-of-function alleles appeared distributed in a single wild population, with the exception of the splicing site mutation widely distributed across The Netherlands. Analysis of the sequence data from the 1001 Genomes Project detected four other putative semidwarf accessions from south Sweden (Sim-1, TV-22, TV-30, and TV-7) that carry the Var 2–1 missense mutation. This result suggests that Var-2 missense loss-of-function allele might be widely distributed at a local scale because Var, Sim, and TV accessions originate from the same south Swedish coastal area (SI Appendix, Table S4).

Genome-Wide Association Study for Plant Height.

Because several of the ga5 semidwarf accessions identified in this study (Tha-1, Sparta, Var 2–1, and T1080) were included in the A. thaliana Hapmap experimental population (15), we tested whether the GA5 locus could be detected by genome-wide association study (GWAS) mapping. Measurements of plant height in 345 accessions of this collection showed a large amount of natural variation and high broad-sense heritability (h_b² = 0.80) (Fig. 3A). However, no marker was significantly associated (P > 0.05 with Bonferroni correction for 214,000 markers; SI Appendix, Fig. S5) with plant height; the largest association was detected on chromosome 4, ∼0.3 Mb away from GA5 (P = 3 × 10⁻⁵; Fig. 3B). Analysis of linkage disequilibrium (LD) showed a complete LD decay 10 kb upstream and downstream of GA5 (Fig. 3C), thus excluding the linkage of the observed association with GA5. By contrast, a significant association was detected when all four GA5 loss-of-function alleles were combined as a single nonfunctional haplotype (P = 2.7 × 10⁻¹⁴). Therefore, despite the strong effect of natural GA5 loss-of-function alleles on plant height, GWAS was unable to detect this locus due to the low frequency of semidwarf accessions and their multiple independent causal mutations.

Fig. 3.

GWAS analyses, population structure, and GA5 diversity. (A) Correlations between height and height up to first silique. Red dots indicate the values from semidwarf accessions. (B) Genome-wide association mapping profile for plant height on chromosome 4. The GA5 position is indicated by an arrow. (C) Linkage disequilibrium 200 kb upstream and downstream of the GA5 locus. The heat color scale represents squared correlation (R²) between pairs of SNPs. (D) Population structure of 100 accessions including nondwarf and GA5 semidwarfs collected in different world regions at K = 5. Colored asterisks indicate accessions carrying different GA5 loss-of-function alleles. (E) GA5 haplotype network. Haplotypes are represented by circles with size proportional to the number of populations containing that haplotype. Each node represents a single mutation.

GA5 Phylogeny and Population Structure.

We determined the genetic relationships among the semidwarf accessions using a structure analysis with 117 genome-wide SNP markers already available (19, 20) or developed in this work. Structure analysis of these accessions found five distinct genetic groups that closely corresponded to the geographic regions of origin of the semidwarf accessions (Fig. 3D and SI Appendix, Fig. S6) in agreement with the strong global geographic structure described in A. thaliana (20). In all cases, semidwarf accessions were genetically more related to the nondwarf individuals from the same population and region than to any other accession, indicating the independent origin and expansion of semidwarfs in these regions. In most populations containing ga5 semidwarfs where five or more individuals were collected, wild-type GA5 alleles were found within the populations, except for the central Asian populations Dja and Sus, in which all individuals were semidwarf. Dja-1 and Sus-1 accessions carried different GA5 loss-of-function alleles (Fig. 3D) regardless of the overall low genetic variation present in central Asia (21). It is also remarkable that different GA5 loss-of-function alleles were found in the Iberian Cat and Mdc populations together with wild-type alleles (Fig. 3D). In contrast, semidwarf genotypes in Dutch populations were very similar and carried the same loss-of-function mutation (SI Appendix, Fig. S7).

Network analysis of the 33 different GA5 haplotypes detected within the genomic GA5 sequence identified a common GA5 functional haplotype which showed a worldwide distribution (Fig. 3E and SI Appendix, Table S5). Twenty other GA5 haplotypes were connected to this frequent haplotype by fewer than five mutational steps and were distributed in all geographic regions. The central network position of the most frequent haplotype suggests that this is the oldest GA5 allele, from which most other haplotypes may have derived by a small number of mutations (Fig. 3E). Furthermore, 14 additional low-frequency haplotypes, which include only Iberian and Asian haplotypes (Cat, Mdc, Mar, Kas, Pak, and Fuk), were separated from the main node of the network by two long related branches. Loss-of-function GA5 haplotypes appeared evenly distributed within this network, and all but one of these alleles were connected by a single mutational step to their presumably ancestral haplotype. In addition, all loss-of-function haplotypes occupied branch-end positions in this network. Therefore, independent GA5 loss-of-function alleles seem to be generated in multiple genetic backgrounds, but they have not produced derived haplotypes (Fig. 3E).

Signatures of Selection at the GA5 Locus.

To estimate the amount and pattern of nucleotide diversity in the GA5 gene, we analyzed the SNP data from 512 accessions available from the 1001 Genomes Project. GA5 shows lower nucleotide diversity within coding regions than introns (Fig. 4A). Total nucleotide diversity (π = 0.0017; SI Appendix, Table S6) was lower than the average nucleotide diversity reported in previous studies [0.0081 for centromeric and 0.0059 for noncentromeric regions (22)]. GA5 also presents a low ratio of nonsilent to silent polymorphism [π(ns)/π(s) = 0.132], which is consistent with a signature of purifying selection, as previously suggested for rice GA biosynthesis genes (23). In addition, significant negative values for Tajima’s D at nonsynonymous sites (D_n) were detected in both the aforementioned 512 accessions (D_n = −2.289, P < 0.01) and the more than 100 accessions used in the present study (D_n = −1.987, P < 0.05), including semidwarf haplotypes (SI Appendix, Table S6). Overall, this pattern is compatible with the occurrence of purifying selection, in which polymorphisms leading to amino acid substitutions are maintained at low frequencies.

Fig. 4.

The GA5 locus shows signatures of natural selection. (A) Nucleotide-sliding window analysis of nucleotide diversity (π) across the GA5 locus in 512 A.thaliana wild accessions. (B) Fay and Wu’s H_n analysis across the GA5 genomic region in populations containing semidwarfs from The Netherlands (blue) and the Iberian Peninsula (red) and populations of normal size (gray). Asterisks denote statistical significance: *P < 0.05 and **P < 0.01.

To test whether positive selection may have contributed to an increase of GA5 loss-of-function alleles, we searched for molecular fingerprints of recent selective sweeps over a region of 80 kb upstream and downstream of GA5 in two populations from two different regions. These Cat (Iberian Peninsula) and OW/Sch (The Netherlands) populations were selected because they contain a moderate frequency of GA5 loss-of-function alleles. One additional population that does not contain semidwarf individuals from each of the regions was analyzed as control. Significant negative values of the normalized Fay and Wu’s H statistics were found around GA5 in the Cat and OW/Sch populations containing semidwarfs (0.019 < P < 0.05) (Fig. 4B), which is consistent with an excess of derived high-frequency mutations that commonly accompanies selective sweeps. We also detected negative values for the Fay and Wu’s H_n statistics in the semidwarf Iberian Peninsula population around the CBF cluster involved in cold acclimation, for which natural variation has been reported (Fig. 4B) (24). This pattern was absent in populations without semidwarfs from the same regions (Fig. 4B). These results suggest that positive selection might contribute to increase the frequency of GA5 loss-of-function mutations under particular environments, although drift and relaxed purifying selection could also contribute to a high frequency of GA5 loss-of-function alleles in some other populations.

Discussion

In this study we have shown that A. thaliana semidwarf genotypes are relatively frequent in natural populations of different regions in the world and mostly caused by mutations in the GA5/GA20ox1 gene. These results evidence a rather simple genetic basis for plant height, but its multiallelic bases hampered GA5 detection by GWAS mapping. GA5 behaves as a functional ortholog of the green revolution genes of rice SD1 and barley Sdw1/Denso. This result points to a conserved evolution for this common trait in crop and wild plant species. Thus, GA 20-oxidase is identified as a hot spot for phenotypic variation in plants (25) and illustrates the usefulness of the analysis of A. thaliana natural variation to find genes of interest for plant breeding. The observation of major phenotypic changes caused by a large number of independent mutations resembles the situation found for the FRIGIDA (FRI) gene of A. thaliana involved in flowering time, another adaptive trait, which indicates that this pattern is not unique but rather common (17). As previously reported for FRI and Flowering Locus C (FLC) flowering genes, most GA5 haplotypes show a subregional or local distribution, but the number of independent functional alleles was significantly larger in the Iberian Peninsula than in northern and central Europe, in agreement with the overall larger Iberian diversity (17, 21, 26).

Our study supports the possible contribution of different evolutionary forces to the occurrence of GA5 loss-of-function alleles in nature. The relatively high frequency of several GA5 loss-of-function alleles in central Asia and within local populations in The Netherlands, central Asia, and Iberian Peninsula suggests an advantage or neutrality. This is especially the case in some populations where multiple mutations have occurred and are still present. The wide geographic distribution of the same GA5 allele found in many locations of The Netherlands separated by more than 100 km indicates that this allele is spreading, further indicating the absence of deleterious effects. In addition, phenotypic characterization of GA5 semidwarf accessions did not detect any strong negative effect on adaptive and fitness traits, which suggests that these alleles do not display any general obvious negative pleiotropic effects or tradeoffs. This result is in agreement with the phenotypes described for artificially induced GA5 loss-of-function mutants, which show similar seed yields to those of wild-type accessions (8). This lack of effect on seed production is probably due to expression of GA20ox paralogs, mainly GA20ox2 (8). Similarly, GA20ox2 mutations in rice and barley do not display tradeoffs (5, 6, 27). By contrast, mutations in early steps of GA biosynthesis have been associated with negative pleiotropic effects, such as the absence of seed germination shown by ga1 null mutants or the reduced fertility and altered flower development observed even in leaky GA1 alleles (13). A similar situation has been reported in rice, where the effects derived from mutations on genes involved in early steps of GA biosynthesis were less favorable for crop production compared with mutations on rice GA20ox2 (28).

Our analyses suggest that both negative and positive selection may act on GA5 loss-of-function alleles. The conditional negative effect of these alleles is suggested by the low frequency of most loss-of-function alleles and by the fact that they are not maintained long enough to derive new haplotypes. Hence, such alleles seem to be transiently maintained in nature. In addition, such a potential negative effect of GA5 loss-of-function alleles is also suggested by purifying selection inferred from the low ratio of replacement to silent polymorphisms and negative Tajima’s D_n values, in agreement with previous reports in rice (23). In contrast, positive selection might contribute to transient increases in the frequency of loss-of-function alleles in certain populations, as suggested by the negative values of Fay and Wu’s H_n tests across the GA5 locus for the two tested populations segregating for semidwarf individuals. Remarkably, this pattern is absent in populations of normal-sized plants from the same regions. Therefore, we reason that allelic variation at GA5 locus might be maintained in nature by antagonist pleiotropy, (i.e., reversed fitness effects in different environments) (29). However, we cannot discard the possibility that GA5 variation shows conditional neutrality in other populations (i.e., loss-of-function alleles might be neutral in some environments but deleterious in others). Neutrality tests should be considered carefully due to the complex demographic history of A. thaliana populations in the wild. Furthermore, the population genetic analysis is agnostic to the local extinction or recolonization dynamics of populations. The identification of signatures for selection using genome-wide screens may be hampered by the occurrence of different loss-of-function GA5 alleles under positive selection, a situation that also affected GWA mapping.

It remains to be determined which are the environmental cues that could contribute to an increase in the frequency of GA5 loss-of-function alleles because these mutations appear distributed in a wide range of anthropoid and natural environments. It has been previously shown that the short-plant height phenotype caused by the well-known erecta loss-of-function mutation provides fitness advantage in static landscapes. On the contrary, the erecta frequency was reduced under disturbed environments (30). Analogously, it can be speculated that environmental stability might favor GA5 semidwarf individuals. Conclusive demonstration about positive, negative, or neutral fitness effects of GA5 loss-of-function alleles depending on the environment will require further analyses under different natural conditions to elucidate the evolutionary forces driving GA5 variation and its ecological significance.

Materials and Methods

Plant Material and Growth Conditions.

Stock numbers and detailed information for accessions used in this work are listed in SI Appendix, Table S1. For allelism tests, semidwarf accessions were crossed with Ler and ga5 (13). To facilitate the allelism tests, male sterility based on the ms1 mutant (31) was introgressed into the ga5 background. Plants were grown under greenhouse conditions at 16 h light, 22 °C/18 °C day/night cycles. For all experiments, seeds were stratified in water at 4 °C for 4–6 d before germination. Ten repetitions per genotype (cross) were conducted. All crossed accessions are listed in SI Appendix, Table S2. The Ooij, Schar, Hoor, Haarl, and Oosth Dutch semidwarf populations and the Mdc Iberian semidwarf population were found in the course of our studies. Allelism was concluded based on sequence data that correlated with the semidwarf phenotypes and haplotypes tested before in allelism tests. Phenotyping for plant height and height up to first silique was conducted 2 wk after flowering because both traits did not change after that date (SI Appendix, Fig. S8). In cases of extreme flowering lateness, plants were vernalized for 6 wk. Flowering time was recorded as days after germination until the first opened flower. Branch number was scored as the number of axillary stems grown from the rosette.

Sequencing of GA5 Gene and Genotyping.

Genomic DNA was isolated from leaf material using the BioSprint workstation (Qiagen). Primers used for GA5 sequencing are detailed in SI Appendix, Table S7. PCR reactions were performed using LA Taq DNA polymerase (Takara) following manufacturer’s instructions. Sanger sequencing of purified PCR products was done by the Max Planck Genome Center Cologne (Cologne, Germany). GenBank accession numbers of DNA sequences generated in this work are listed in SI Appendix, Table S5. SNP genotyping of new accessions collected in this study was done as described in previous works (19, 20) using the genotyping facility service of the University of Chicago (Chicago).

Statistical Analysis.

Descriptive statistics, t tests, Tukey tests, and principal component analyses were conducted with R. The method of efficient mixed-model association expedited was used for GWAS (32) using kinship matrix to correct for population structure. LD analysis was performed with the R package LD heatmap (33).

Structure Analysis.

Population structure was inferred using model-based clustering algorithms implemented in the software STRUCTURE, using the haploid setting and running 20 replicates with 50,000 and 20,000 MCMC (Markov chain Monte Carlo) iterations of burn-in and after-burning length, respectively (34). To determine the K number of significantly different genetic clusters, we applied the ΔK method in combination with the absolute value of ln P(X|K) (35).

Population Genetics: Fay and Wu’s H Statistics and Haplotype Network.

Population genetics analyses were conducted with the software DnaSP (DNA Sequence Polymorphism version 5.10) (36). The normalized Fay and Wu’s H was performed as described (16) in populations containing semidwarfs from The Netherlands and the Iberian Peninsula (SI Appendix, Table S8). Representative accessions of different populations from central Spain, with no semidwarfs, were used as control (SI Appendix, Table S8). For the Dutch control population, accessions from a rural area northeast of Wageningen were collected with no prior knowledge of semidwarfness occurring in this population (SI Appendix, Table S8). The sequences of GA5 (At4g25420) and flanking genes were obtained after specific PCR amplification from genomic DNA and sequencing using Applied Biosystems 3730XL automated sequencers (SI Appendix, Table S7). Sequences were aligned with ClustalW (37) and manually inspected. Arabidopsis lyrata sequences were obtained by BLAST search (www.phytozome.net) and used as out-group to assign ancestral and derived states to SNP variants. To assess the statistical significance of Fay and Wu’s H, we computed 10,000 coalescent simulations in DnaSP version 5.10 (36). The haplotype network of GA5 was constructed using TCS1.21 (38) that implements a maximum parsimony method and excluding gaps as events in the analysis. Insertions and deletions in the semidwarf accessions were considered as single events and added manually to the haplotype network.