Bimodal expression level polymorphisms in Arabidopsis thaliana

Abstract

Differences in gene expression are termed expression level polymorphisms (ELPs). Here, we propose a new ELP class, bimodal ELPs (bELPs), as a criterion to screen for genes that are responsible for natural phenotypic variation and/or that are targeted by balancing selection. bELP genes are characterized by two expression level modes. Genomic scans based on nucleotide sequences are not ideal for identifying genes targeted for selection. A critical concern is that several genes can be present in the selection-targeted regions identified by such scans. This situation indicates the importance of integrating genomic sequence data and other information, such as gene expression data. Comparative transcriptomics is useful for determining evolutionarily and ecologically important polymorphisms. In a genome-wide expression screen of 34 accessions, we identified 344 Arabidopsis thaliana genes exhibiting bELPs. Population genetic analysis revealed that bELP genes had high nucleotide diversities and long linkage disequilibriums. The highest nucleotide diversity (11-fold greater than the genomic mean) was found in the At1g23780 gene, which encodes a putative F-box protein. We observed a clear association between the expression mode and sequence type of the At1g23780 gene. Our results suggest that bELPs will be useful for the screening and functional analysis of genes responsible for phenotypic polymorphisms. Such a “multi-omics” approach has the potential to facilitate the scanning of genes relevant to balanced polymorphisms not only in A. thaliana, but also in other model and non-model organisms.

Introduction

Balanced polymorphisms are characterized by two or more alleles that are selectively maintained within populations or species.^1-3 Several selective mechanisms that can maintain polymorphisms have been proposed, such as frequency-dependent selection, spatially and temporally unstable selection and heterozygote advantage.² Balanced polymorphisms are associated with specific levels and patterns of nucleotide variation at the selective target and linked genomic regions. Thus, the molecular signature of a given locus can be used to detect an adaptation. This signature can include intermediate frequency polymorphisms and a peak of increased nucleotide diversity.⁴ Molecular evidence for balanced polymorphisms has been obtained for several genes and/or genomic regions, e.g., the human class I and II MHC genes,⁵ Drosophila Adh,^6,7 Arabidopsis thaliana MAM⁸ and disease resistance loci.^9-12

Genomic scans based on nucleotide sequences are used to identify genes targeted by selection. Even though nucleotide polymorphisms are the primary genetic cause of phenotypic variations, these genomic scans have a drawback in that numerous genes can be present in selection-targeted regions identified by genomic scans. A gene targeted by selection cannot be identified from other candidate genes except in cases where a clear peak of nucleotide diversity is observed, because linkage disequilibrium (LD) and/or a stretch of high nucleotide diversity spans a large genomic region that contains many candidate genes.^13,14 In such cases, additional information, such as expression data, is needed to determine the true selection target. Expression data provides clues about functional polymorphisms as expression level polymorphisms (ELPs). ELPs are defined as differences in gene expression between individuals.^15,16 The levels of ELPs in a genome macroscopically correlate with the levels of nucleotide diversity.¹⁵ ELPs have been shown to contribute to inter- and intra- specific phenotypic variation.¹⁷ Examples of such variation in plants include the liberation of the kernel from the hardened casing, as observed in maize domestication,¹⁸ the loss of seed shattering in rice domestication,¹⁹ flowering time control,^20-22 pathogen resistance,^23,24 insect resistance and secondary metabolism in Arabidopsis.^25-27 ELPs can be caused by various types of DNA sequence polymorphisms, including those affecting trans-acting factors, cis-acting promoter regions, cis-acting splice sites and whole or partial gene deletions.

Recent advances in molecular genetic techniques make it possible to identify loci responsible for adaptation and to study how these loci are selected.^28-30 In particular, identifying genes targeted by balancing selection and unveiling the genetic mechanisms that give rise to natural variation are important for both basic science (e.g., the functional analysis of biological mechanisms) and applied science (e.g., molecular breeding assisted by genomics).³¹ In most previous studies, the genes responsible for natural phenotypic variation and/or the genes that are targeted by balancing selection have been identified by a combination of genetic mapping (e.g., QTL mapping and positional cloning) and functional molecular analysis of candidate genes (e.g., expression analysis and transgenic assay).^32,33 This situation indicates the importance of integrating multiple data from various experiments.

In this article, we propose a new class of ELPs, bimodal expression level polymorphisms (bELPs), as a criterion for screening for candidate genes targeted by balancing selection. bELP genes are characterized by two distinct expression levels. In other words, two modes are distinguishable in a histogram of expression levels of bELP genes (Fig. 1A). A bELP is an associated concept with a large effect eQTL (expression QTL); some of the bELPs found in natural populations may also be identified by quantitative genetics using mapping populations, and vice versa.^34-36 We have identified genes that exhibit bELPs in the A. thaliana genome. Sequencing and population genetic analysis of bELP genes revealed that they have significantly greater nucleotide diversity and longer blocks of linkage disequilibrium than the genomic average. Thus, regions containing bELP genes exhibit characteristics of balancing selection. Our results suggest that a “multi-omics” approach, namely the integration of bELP screening with evolutionary population genomics, will provide a good starting point to conduct a further functional analysis of balancing selection.

Figure 1. Schematic representation of bELPs. (A) Histograms depicting typical ELPs (left) and bELPs (right). In the bELP histogram, two expression modes are clearly distinguishable. The ELP gene superclass includes the class of bELP genes. (B) Screening criterion for identifying bELP genes. Expression levels of a specific gene in each accession are plotted (circles) on an axis. d_max is the greatest difference observed between the expression levels of two neighboring accessions.

Results

Screening for bELPs in the A. thaliana genome

The A. thaliana genome was first screened for genes exhibiting bELPs using a publicly available DNA microarray data set (see Materials and Methods). At first, we tried to fit the expression data of each gene to a mixture distribution of two normal distributions (Fig. S1). However, we realized that it was difficult to obtain reasonable results using this approach, because of insufficient data to estimate the distribution. Thus, we decided to employ a simple non-parametric strategy for the screening (see Materials and Methods). Using lower thresholds, more genes were identified as bELP genes (103 to 872 genes, Table 1). The ratio of singletons (71% to 73%) did not depend on the threshold. We used d_max = 1 as the threshold for the following analysis, because it corresponds to a 2-fold change, which is often designated as a threshold in empirical studies using DNA microarrays. A 2-fold difference between levels of expression was expected to be clearly distinguishable when using the DNA microarray and real-time PCR techniques. With this threshold, 344 genes were identified as bELP genes (Fig. 2A; Fig.S2 and TableS3). The frequency of two expression types varied from 2:32 to 17:17 (Fig. 2B). To assess whether sequence polymorphisms on the probe positions affect the apparent expression level, we sequenced one of the bELP genes, At1g23780, in 22 accessions (Fig.S3). We identified 13 SNPs on 7 of 11 probe positions. The exact same set of SNPs was found in both the high and low expression type. This showed that the SNPs on the probe positions did not affect the apparent expression level of the At1g23780 gene. In addition to this result, it was estimated that SFPs (single feature polymorphisms) were responsible for ~0.13% of the apparent variance in expression level in the previous study using Col-0 and Cvi-1.¹⁵ Therefore, we concluded that most of the bELPs were true ELPs, and not the result of mis-hybridization by sequence polymorphisms. The bELP genes included a large number of biotic stress-related genes (GO: 0030383 host-pathogen interaction, p = 3.4 × 10⁻⁵; and GO: 0003793 defense immunity protein activity, p = 1.0 × 10⁻⁴) and genes whose products are located in endomembrane systems (GO: 0012505 endomembrane system, p = 2.8 × 10⁻⁵). The bELP genes included some genes known to be targets of balancing selection, e.g., ESP and At1g63880 (Fig. 2C), as detailed in the Discussion.

Table 1. Results of bELP screening with various thresholds.

Threshold		bELP
Log2	Fold change	With singleton	Without singleton	Ratio of singleton (%)
0.58	1.5	2982	872	71
1	2	1276	344	73
1.58	3	580	165	72
2	4	370	103	72

Figure 2. bELPs in A. thaliana and analyses of the selected sequences. (A) Heat map of bELP gene expression (dmax = 1). Low and high levels of expression are shown by blue and yellow colors, respectively. Thirty-four accessions are aligned horizontally and 356 genes are aligned vertically. Red arrowheads indicate the 20 sequenced genes. A full resolution image with AGI codes and annotations of bELP genes is available as Figure S2. (B) Frequency distribution of the ratio of two expression types. Inset, enlarged view. (C) Examples of bELP genes. Each histogram shows two expression modes. (D) Histogram of silent-site nucleotide diversity (π) of the 20 sequenced bELP genes (red, upper) compared with the genomic distribution (gray, two lower histograms). Data for the genomic distribution are from Schmid et al.³⁹ and Nordborg et al.³⁸ (E) Histogram of Tajima’s D of the 20 sequenced bELP genes (red, upper) compared with the genomic distribution (gray, lower). Data for the genomic distribution are from Nordborg et al.³⁸ (F) Histograms of the lengths of LD blocks. Upper: LD blocks including bELP genes. Lower: genome-wide distribution estimated from Clark et al.⁴⁰ LD blocks over 250 kb (0.002%) are not shown in the lower histogram.

Population genetic analysis of 20 bELP genes

To determine the level of nucleotide diversity (π) of bELP genes, genomic sequences of 20 bELP genes were obtained from eight to 23 accessions. Upstream regions and 5′UTRs of the genes were sequenced, because these regions may contain nucleotide polymorphisms responsible for bELPs. We obtained a total of 210,824 bp of nucleotide sequences and identified 526 non-redundant polymorphic sites (Table 2). Silent-site nucleotide diversities of the genes were highly variable (π = 0.0005 ~0.0782; Fig. 2D and Table 2). Thirteen of the 20 sequenced bELP genes had greater silent-site nucleotide diversity than the mean level of 0.007 observed from previously studied A. thaliana genes.³⁷ A comparison of the distributions of the silent-site nucleotide diversity of bELP genes with that of the random genome-wide data set^38,39 suggested that bELP genes exhibited higher silent-site nucleotide diversity (Fig. 2D, p < 0.01, Mann-Whitney’s U test). To assess whether high silent-site nucleotide diversity is a specific feature of bELP genes or a general feature of ELP genes, 344 genes exhibiting the most variable expression (but not exhibiting bELP) were collected and compared with bELP genes. As a result, we found that the bELP genes had significantly greater silent-site nucleotide diversity than did genes exhibiting expression polymorphisms but not bELPs (p < 0.01, Mann-Whitney’s U test). However, with respect to Tajima’s D statistics, there was no significant positive departure from a neutral-equilibrium model, although some genes exhibited significantly negative Tajima’s D values (Table 2). The distribution of Tajima’s D of bELP genes was not significantly different from that of the random genome-wide data set (Fig. 2E).

Table 2. Measures of diversity for the 20 bELP genes.

AGI code	na	Lengthb (bp)	Sc	πd	θwd	Tajima's D	Fu and Li's D*
At1g11280	17	610	10	0.0024	0.0049	-1.8776*	-2.6346*
At1g23780	23	577	114	0.0782	0.0569	1.5053	0.9226
At1g52040	22	419	44	0.0157	0.0255	-1.6706	-2.3791
At1g53690	23	544	24	0.0089	0.0121	-0.9712	0.4331
At1g58270	21	725	50	0.0227	0.0228	-0.1620	-0.0944
At1g64190	21	549	36	0.0246	0.0194	0.9314	0.4395
At1g73330	23	544	3	0.0005	0.0015	-1.7313	-2.4937
At2g25450	8	432	13	0.0091	0.0127	-1.4567	-1.6851
At2g40010	17	516	29	0.0083	0.0183	-2.2182**	-3.0134*
At3g27200	20	474	31	0.0165	0.0187	-0.5667	-0.4396
At3g26744	15	642	28	0.0102	0.0136	-1.0486	0.0394
At3g44280	12	592	13	0.0059	0.0077	-0.9857	-1.3601
At4g03060	13	644	5	0.0013	0.0027	-1.8631*	-2.3235*
At4g05050	24	700	40	0.0173	0.0170	0.0709	-0.1278
At4g22380	21	618	24	0.0115	0.0119	-0.1277	0.0010
At4g23810	15	539	6	0.0021	0.0035	-1.3326	-1.9560
At5g01820	20	454	19	0.0128	0.0125	0.0983	0.4566
At5g10400	18	498	13	0.0065	0.0084	-0.8726	-1.1086
At5g44520	17	600	39	0.0133	0.0195	-1.3134	-2.0006
At5g53940	22	603	21	0.0078	0.0097	-0.7315	-0.5220

^**p < 0.01, *p < 0.05; ^anumber of samples; ^blength of sequenced region; ^cnumber of segregation sites in the sample; ^destimates are based on silent site.

Population genetic analyses based on previously published array-based resequence data

In addition to the analyses of 20 bELP genes, we conducted further population genetic analyses based on the genome-wide polymorphism data of 20 accessions of A. thaliana.⁴⁰ First, we compared the lengths of the LD blocks including a bELP gene to the genome-wide distribution of the LD block lengths (Fig. 2F). The lengths of the LD blocks that included a bELP gene were highly variable (953 bp ~467,183 bp) and significantly longer than those of the genome-wide data (Mann-Whitney’s U test; p < 2.2 × 10⁻¹⁶). We found that 92.2% of bELP genes were located on longer LD blocks than the median LD block length of genome-wide distribution (3,797 bp).

The π of 4-fold degenerate sites (50-kb window) (provided by Clark et al.⁴⁰) in regions that included bELP genes were compared with those distributed in genome-wide regions. Genomic regions, including bELPs, exhibited significantly greater π values of 4-fold degenerate sites than the genome-wide data (Fig. S4; Mann-Whitney’s U test; p < 0.0045).

Sequencing analysis of At1g23780 and flanking regions

One of the sequenced 20 bELP genes, At1g23780, exhibited an 11-fold greater silent-site nucleotide diversity (π = 0.0782) than the genomic mean level. Although a high nucleotide diversity level is a signature of balancing selection, genetic hitchhiking may cause this diversity. To assess whether the At1g23780 gene was a target of the balancing selection, we performed further sequencing analysis that included the flanking genes (At1g23760 ~At1g23800). We obtained a total of 128,472 bp of nucleotide sequences of these five genes from 23 A. thaliana accessions and from three Arabidopsis lyrata individuals. For A. thaliana accessions, 292 non-redundant polymorphic sites were determined (Table 3). Nucleotide diversities were calculated with a sliding window of 100 bp, in steps of 25 bp. The levels of nucleotide diversity from At1g23780 to At1g23800 were greater than the genomic mean level (Fig. 3A). A peak of nucleotide diversity was found at the At1g23780 locus and its upstream region, suggesting that the At1g23780 gene is a target of balancing selection. Neighbor-joining trees of the genes show that At1g23780, At1g23790 and At1g23800 fell into two major clades (Fig. 3B). The frequencies of the sequence groups were at intermediate levels. LD analyses based on the polymorphism data around At1g23780 and its flanking region (~35 kb), provided by Clark et al. (2007), showed that there is a high LD region between At1g23780 and At1g23800. These features indicate the presence of balancing selection involving the region containing At1g23780–800. Although At1g23790 and At1g23800 had two sequence groups, these genes had only one expression mode (Fig. 3C). Only At1g23780 exhibited bimodal expression. The expression modes and sequence groups of At1g23780 were correlated, except for two accessions (Van-0 and Kin-0 in Fig. 4A). The At1g23780 gene sequence of Van-0 is a chimera of Col-0-type and Cvi-type sequences. Single nucleotide polymorphisms (SNPs) among Col-0, Cvi and Van-0 are shown in Figure 4B. In the Van-0 accession, Col-0-type SNP alleles formed a cluster (p < 1.4 x 10–23). This result suggests that the sequence of Van-0 arose by two recombination events or by gene conversion. The promoter sequence of Van-0 was Cvi-type, although Van-0 exhibited a higher level of expression than Col-0. The recombinant region in Van-0 might affect its expression level. For Kin-0, we could not identify sequence differences that explained the differences in expression. A possible explanation is the existence of a trans-factor polymorphism. Polymorphism of a transcriptional factor controlling expression of At1g23780 might explain the expression level polymorphism in Kin-0. The product of the At1g23780 gene is annotated as a putative F-box protein. Because amino acid substitutions were not found in the F-box domain, the At1g23780 protein was expected to act as an F-box protein in every accession that we sequenced.

Table 3. Measures of diversity for the genes flanking At1g23780.

AGI code	na	Lengthb (bp)	Sc	πd	θwd	Tajima's D	Fu and Li's D*
At1g23760	23	590	5	0.0013	0.0041	-1.81765*	-2.40051
At1g23770	23	555	9	0.0059	0.0066	-0.72468	-0.7821
At1g23780	23	2,820	238	0.0485	0.0341	1.71137	1.03643
At1g23790	23	589	19	0.0218	n.a.	1.56202	0.70176
At1g23800	23	585	21	0.0290	0.0168	2.57462**	0.79733

^**p < 0.01, *p < 0.05; ^anumber of samples; ^blength of sequenced region; ^cnumber of segregation sites in the sample; ^destimates are based on silent site.

Figure 3. Nucleotide diversity, neighbor-joining trees and expression levels of At1g23780 and flanking genes. (A) Nucleotide diversity was calculated with a sliding window (100 bp). Gray arrows indicate genes. The dashed line indicates the average level of silent-site π for A. thaliana. (B) Neighbor-joining trees of At1g23780 and flanking genes (from At1g23760 to At1g23800). Open circles and closed triangles indicate low and high expression accessions of the At1g23780 gene, respectively. Closed boxes represent A. lyrata as an outgroup. (C) Gene expression levels of At1g23780 and flanking genes. Data for At1g23770 are not available, because there was no corresponding probe-set on the ATH1 gene chip. (D) The r² plot of At1g23780 and its flanking region (~35 kb). Black indicates strong LD (r² = 1) and white indicates weak LD (r² = 0).

Figure 4. Relationship between haplotypes of the At1g23780 gene. (A) The haplotype network for At1g23780. Names of accessions with low expression levels are shaded. Minimum steps are indicated for each node. (B) Col-0 and Cvi polymorphic nucleotides. Nucleotides also shared by Col-0 are depicted by dots. Nucleotides also shared by Cvi are shaded. (C) Schematic of the At1g23780 gene. The upper line indicates the putative recombined region in the Van-0 accession. Triangles indicate amino acid substitutions. (Closed triangles represent amino acid differences between the Col-0 and Cvi alleles; gray triangles indicate minor variations). The closed box indicates the F-box domain.

Discussion

We found that the silent-site nucleotide diversity of bELP genes is significantly greater than that of the random genome-wide distribution (Fig. 2D). A greater nucleotide diversity is expected when the polymorphisms are maintained for long periods of time, which is considered to be a possible signature of balancing selection.⁴ We also found that the lengths of the LD blocks that included bELP genes were significantly longer than those in the genome-wide distribution (Fig. 2F). A longer LD is expected when balancing selection occurs, e.g., a recent, partial selective sweep associated with local adaptation.⁴ Our results suggest that balancing selection is occurring on the genomic regions containing bELP genes. bELPs could be useful for the screening of candidate genes targeted by balancing selection. The association between bELPs and genomic signatures of balancing selection suggests that the majority of bELPs are the result of cis-acting nucleotide polymorphisms (e.g., polymorphism of cis-acting regulatory region, splicing junction and whole gene deletion). This is consistent with previous reports of ELPs and of the molecular basis controlling ELPs.^15,41

The bELP genes identified in this study include genes that were previously studied as targets of balancing selection. An example is ESP (Table S3), the product of which promotes herbivory by generalist herbivores and decreases oviposition by specialist herbivores. ESP is thought to be selected depending on the surrounding herbivore fauna.^27,42 Another example is the At1g63880 gene, which encodes a putative TIR-NBS-LRR-type disease resistance protein (Fig. 2C). At1g63880 is located in the high diversity region 1 defined in Cork et al., 2005, which was identified by analysis of a large-scale sequence data set. The genomic region around At1g63880 exhibited high nucleotide diversity (π = 0.068). The TIR-NBS-LRR gene cluster containing At1g63880 was suggested to be a target of balancing selection,¹³ an example that suggests that ELP analysis helps to isolate selection targets from among candidate genes identified by genomic scans of nucleotide sequences.

The physiological roles of some bELP genes have been examined in previous studies. However, the modes of selection for such genes were not addressed. For example, inducer of CBF Expression 1 (ICE1) is a key transcriptional factor for controlling resistance to cold stress.^43-45 The expression of ICE1 exhibited a clear bELP profile (Fig. 2C). This result raises the possibility that ICE1 has a role in phenotypic polymorphisms with respect to cold resistance. The ICE1 gene was also reported as the SCREAM (SCRM) gene, which determines successive initiation, proliferation, and terminal differentiation of stomatal cell lineages.⁴⁶ This surprising finding suggests a link between cold resistance and stomatal development. The pleiotropic function of ICE1 may play a role in maintaining its bELP status.

Our list of bELP genes includes some known to be responsible for phenotypic polymorphisms (i.e., FLC, ESP, AOP2, AOP3 and RPP5). This result suggests that bELP analysis is effective in screening for genes associated with phenotypic polymorphisms. However, our list does not represent all of the genes responsible for phenotypic polymorphisms. One possible reason for the omission of some genes is that the phenotypic polymorphism is not derived from ELPs but from changes in protein function caused by amino acid substitutions. It is unclear to what degree balancing selection results from ELPs. A combination of large-scale resequence analysis⁴⁰ and systematic multi-conditional ELP analysis will reveal the quantitative importance of ELPs for balancing selection.

It is worth mentioning that the bELP genes include several genes responsible for fundamental cellular processes, e.g., histones, actin, tubulins, ubiquitin and ribosomal proteins (Fig. 2C; Table S3). These genes are expected to have a stable expression level among accessions, because they function in fundamental cellular processes and are ubiquitously expressed in various tissues and at different developmental stages. One possible reason is that they are present in multiple copies, which may compensate for the expression level of a specific polymorphic gene. Another possible reason is selection for unknown pleiotropic functions of the genes. These genes are commonly used as internal standards in expression analyses, because their expression is thought to be stable. This result suggests that care should be taken in selecting internal standard genes.

The At1g23780 gene was identified as a possible target of balancing selection. This gene encodes a putative F-box protein, a subunit of E3-type ubiquitin ligases, which are key components for defining which proteins are degraded by proteasomes. Protein degradation by proteasomes controls various biological processes, for example hormone reception, signal transduction and the defense response (LECHNER et al., 2006). It is not known in which biological processes the At1g23780 protein plays a role. Unfortunately, we could not obtain a knockout line of the At1g23780 gene. In addition to At1g23780, the gene family encoding F-box proteins exhibits high nucleotide diversity.⁴⁰ F-box proteins seem to be good candidates for the functional analysis of polymorphisms at the molecular level.

We found that bELP genes were generally located on genomic regions with significantly longer LDs than the genomic distribution. This result indicates that a flanking region of each bELP gene has a population genomic pattern similar to that of the respective bELP gene. Polymorphisms targeted by balancing selection in long LD regions are hard to identify by scans based on genomic sequences alone. In such cases, functional analysis of the balancing selection is needed to elucidate the molecular mechanisms underlying the selection. Which gene is involved in the selection? Which polymorphisms are involved in the selection? How do the polymorphisms affect the phenotype targeted by selection? We believe that ELPs provide good starting points for further functional analysis, because many ELPs are reported as causes of phenotypic variations^19,24,25,27 and because a measurement of expression level is usually easier to make than a biochemical and cell biological analysis of protein function. Indeed, we found a similar gene genealogy, a high nucleotide diversity and a linkage block from At1g23780 to At1g23800 (Fig. 3A, B and D). Only At1g23780 showed a clear bELP pattern in this region (Fig. 3C). While identifying ELP is not conclusive evidence that the gene is the target of selection, it will be worth conducting a further functional analysis of At1g23780. Such “multi-omics” approaches have the potential to facilitate the scanning of genes relevant for balanced polymorphisms not only in A. thaliana but also in other model and non-model organisms.

Materials and Methods

Screening for bimodal expression level polymorphisms

Expression data were obtained from Detlef Weigel’s website (www.weigelworld.org/resources/microarray/AtGenExpress/). The data were generated as a part of the AtGenExpress project and were also deposited in TAIR as ME00374 AtGenExpress: Ecotypes (triplicated data from 10 accessions) and ME00375 AtGenExpress: Ecotypes singletons (non-replicated data from 24 accessions).²² Triplicate ME00374 data were averaged and used for fitting a mixture distribution of two normal distributions and screening by the simple nonparametric approach. Data were normalized to the average expression levels of each accession. ANOVA was conducted using normalized data of ME00374. The correction of multiple testing by ANOVA was conducted by controlling the false discovery rate (FDR = 0.01).⁴⁷ The estimation of a mixture distribution of two normal distributions from expression data was conducted using the vglm (family = mix2normal) function in the VGAM package of R.^48,49 In the simple nonparametric approach, bELPs were screened with bimodal3.pl, a Perl script developed in-house. Procedures to assess whether a given gene was a bELP gene are described below. First, the expression values of each accession were ordered and a progression of differences was calculated. Next, the maximum value for the progression was designated as d_max (Fig. 1B). Finally, d_max was compared with a threshold. If the d_max was larger than the threshold, the gene was treated as a bELP gene. Based on the screening method, the term “bELP” was defined in this study as a gene that exhibits at least two modes of expression. The Perl script for screening is available upon request. The statistical analysis of the gene ontology was performed with GeneSpring GX (Agilent Technologies).

Plant materials and growth conditions

A. thaliana accessions (Table S1) were from Europe, North America, Central Asia and North Africa. A. thaliana seeds were obtained from ABRC. Arabidopsis lyrata (W313) seeds were provided by Kentaro K. Shimizu (University of Zurich). Seeds were surface-sterilized and then sown onto 0.5% (w/v) Gellan Gum (Wako) with MS medium (Wako) and 1% (w/v) sucrose and grown at 22°C under continuous light conditions.

Sequencing

Genomic DNA was isolated from the aerial parts of three A. lyrata plants and 23 A. thaliana accessions. PCR primers were designed from Col-0 genomic sequences (AGI 2000) using Primer3 (ROZEN and SKALETSKY 2000). Primer sequences are described in Table S2. Some A. lyrata primers were designed from the A. lyrata sequences generated in this study (Table S2). PCR was performed with ExTaq DNA polymerase (Takara), and amplified PCR products were purified using MagExtractor (TOYOBO). A. thaliana sequences were determined by direct sequencing of amplified PCR products. Purified PCR products of the A. lyrata sample were cloned into the T7Blue vector (Merck, Darmstadt) using the DNA Ligation Kit ver.2 (Takara). Plasmids were isolated using QIAprep (Qiagen). At least four independent clones were sequenced. Sequence data from this article have been deposited into the DDBJ/EMBL/GenBank Data Libraries under accession nos. AB498072-AB498554.

Molecular population genetic analysis and phylogenetic analysis

Sequences were initially aligned using Clustal W.⁵⁰ The alignment was manually checked and corrected. Calculations of population genetic parameters were performed with DnaSP 4.10.9.⁵¹ Silent-site nucleotide diversity was estimated as both π⁵² and θ_w.⁵³ Tajima’s D⁵⁴ and Fu and Li’s D^*55 were also estimated. Phylogenetic analysis was performed using PAUP4.0*b10⁵⁶ to decipher the relationship between the ELPs of the highly polymorphic gene At1g23780 and its nucleotide sequence. The empirical genomic distributions of π and Tajima’s D in the silent site and intergenic regions were estimated from published sequence data.³⁸ We estimated π and Tajima’s D of populations consisting of A. thaliana accessions that were used in this article for bELP screening and that were sequenced in Nordborg et al.³⁸ While almost all of the genes described in Nordborg et al.³⁸ were used for estimation of π and Tajima’s D, the following nucleotide sites were excluded: those with alignment gaps and missing data, and all sites in gene 22387873 on chromosome 1, because the number of nucleotides in the sequence file and in the annotation file were mismatched.

Linkage disequilibrium and recombination analysis

For LD analyses, polymorphism data provided by the Perlegen oligonucleotide array of 20 accessions was used.⁴⁰ While Clark et al.⁴⁰ proposed several algorithms for SNP detection, SNPs called “MBML2” were used in our analyses. The following sites were excluded from the analyses: SNPs containing at least one accession that was not detected by the “MBML2” and SNPs that were singletons. The length of an LD block was defined as the distance between the first flanking SNP on the downstream side of the focal polymorphism that fails the four-gamete test⁵⁷ and the focal polymorphism. Note that block length based on the four-gamete test creates a bias for a relatively longer estimate, because the four-gamete test is an estimate of the minimum number of recombinations. The r2 plot of At1g23780 and its flanking region was generated by Haploview 4.0.⁵⁸ Several bELP genes were excluded from the analyses. These were unannotated genes and genes located at the beginning or the end of a chromosome, where there were too few markers to estimate the lengths of the LD blocks. The histogram presented in Figure 2F was developed not on a gene-wise basis, but on a block-wise basis, because sometimes more than one bELP gene was found on the same LD block. Finally, 306 out of 344 bELP genes (Fig. 2F) were used for making comparisons of distribution by the Mann-Whitney’s U-test.