DOI: 10.1534/genetics.107.083279
Low Levels of Polymorphism in Genes That Control the Activation of
Defense Response in
Arabidopsis thaliana
Erica G. Bakker,*
,1M. Brian Traw,
†Christopher Toomajian,
‡Martin Kreitman*
and Joy Bergelson*
,2*Department of Ecology and Evolution, University of Chicago, Chicago, Illinois 60637,†Department of Biological Sciences, University of Pittsburgh, Pittsburgh, Pennsylvania 15260 and‡Department of Molecular and Computational Biology,
University of Southern California, Los Angeles, California 90089 Manuscript received October 10, 2007
Accepted for publication December 21, 2007
ABSTRACT
Plants use signaling pathways involving salicylic acid, jasmonic acid, and ethylene to defend against pathogen and herbivore attack. Many defense response genes involved in these signaling pathways have been characterized, but little is known about the selective pressures they experience. A representative set of 27 defense response genes were resequenced in a worldwide set of 96 Arabidopsis thalianaaccessions, and patterns of single nucleotide polymorphisms (SNPs) were evaluated in relation to an empirical distribution of SNPs generated from either 876 fragments or 236 fragments with.400 bp coding sequence (this latter set was selected for comparisons with coding sequences) distributed across the genomes of the same set of accessions. Defense response genes have significantly fewer protein variants, display lower levels of non-synonymous nucleotide diversity, and have fewer nonnon-synonymous segregating sites. The majority of defense response genes appear to be experiencing purifying selection, given the dearth of protein variation in this set of genes. Eight genes exhibit some evidence of partial selective sweeps or transient balancing selection. These results therefore provide a strong contrast to the high levels of balancing selection exhibited by genes at the upstream positions in these signaling pathways.
P
LANTS defend themselves in response to pathogens and parasites, ranging from viruses to insect her-bivores, using both physical and chemical defenses. Chemical defenses include the synthesis of antimicrobial compounds, such as phytoalexins and defensins, and the production of pathogenesis-related (PR) proteins, com-prising enzymes capable of degrading pathogen cell walls. Defense responses often begin with a gene-for-gene recognition of an avirulence (avr) gene of the pathogen by a resistance (R) gene of the host (VanDerBiezen and Jones 1998; Dangl and Jones 2001; Staskawiczet al.2001).Rgene-mediated resistance is usually accompanied by reactive oxygen species (ROS) and hypersensitive cell death (HR) (Veronese et al. 2003), which can be followed by activation of a salicylic acid (SA)-dependent signaling pathway (Figure 1) and the consequent expression of a characteristic set of defense genes (Glazebrook2001). As a result, plants become more resistant to subsequent attack by otherwise virulent—in the sense of ‘‘ability to infect’’—pathogens, a phenomenon called systemic acquired resistance (SAR) (Sticheret al.1997; Durrantand Dong2004). The SA signaling pathway is activated not only as a result of R gene action but also as a result of pathogen-derived signals. Additional plant defense responses include those controlled by signal transduction networks re-quiring jasmonic acid ( JA) and ethylene (ET). The JA/ ET-dependent signaling pathways mediate resistance responses to herbivores or necrotrophic pathogens (Thommaet al. 2001), whereas the SA-dependent sig-naling pathway mediates resistance to biotrophic patho-gens (Ryalset al. 1996). The SA, JA, and ET signaling pathways interact extensively (Glazebrook2001; Traw
et al.2003). Finally, gene-for-gene interaction between anRgene and its correspondingavrgene in the patho-gen (Flor1956) can trigger downstream pathways in addition to SAR, of which three have been identified Sequence data from this article have been deposited with the EMBL/
GenBank Data Libraries under the following accession numbers: At1g08720, EU404192–EU404286; At1g52400, EU404287–EU404376; At1g54040, EU404377–EU404472; At1g64280, EU404473–EU404568; At2g37040, EU404569–EU404664; At3g11660, EU404665–EU404760; At3g20770, EU404761–EU404856; At3g27920, EU404857–EU404951; At3g44880, EU404952–EU405047; At3g48090, EU405048–EU405143; At3g57260, EU405240–EU405333; At4g31500, EU405334–EU405429; At4g37000, EU405430–EU405525; At4g39030, EU405622–EU405717; At5g06950, EU405718–EU405813; At5g45110, EU405814–EU405908; At5g51700, EU405909–EU406003; At5g64930, EU406004–EU406099; At1g66340, EU406100–EU406195; At4g26120, EU406196–EU406291; At5g13160, EU406292–EU406386; At2g29980, EU406387–EU406482; At2g39940, EU406483–EU406578; At3g26830, EU406579–EU406674; At5g15410, EU406675–EU406770.
1Present address:Department of Horticulture, Oregon State University,
Corvallis, OR 97331.
2Corresponding author:Department of Ecology and Evolution, University
of Chicago, 1101 E. 57th St., Chicago, IL 60637. E-mail: jbergels@midway.uchicago.edu
(Aartset al.1998; McDowellet al.2000; Glazebrook 2001).
Defense response signal transduction pathways in-clude many different genes that together control the activation of defense response as a reaction to pathogen attack and will therefore be referred to as ‘‘defense response genes.’’ There is presently no information re-garding the selective pressures that act on this group of genes in plants. Evolutionary analyses on receptor molecules in the innate immune system of insects have produced diverse results. In general it appears that bal-ancing selection as observed for MHC genes and plant R genes is uncommon for innate immunity genes, al-though in some instances there are records of transient forms of balancing selection (Lazzaro2005). In Dro-sophila, a fruit fly, there is substantial evidence for adap-tive evolution in a suite of genes involved in recognition of pathogens and upstream humoral response signaling pathways, corresponding with an arms race between hosts and parasites (e.g., Relish in Drosophila simulans, Schlenkeand Begun2003; Littleet al.2004; Jiggins and Kim2006, 2007). In many other instances however, Drosophila innate immunity genes appear to be under purifying selection. The end products of response signaling pathways are antimicrobial peptides that pro-vide a generalized pathogen resistance response. Se-quence analysis of antimicrobial peptides produced by D. melanogasterfail to reveal evidence of recurrent direc-tional selection, but instead reveal typical levels of stand-ing silent variation and low levels of divergence across species (Lazzaro and Clark 2003; Jiggins and Kim 2005). Purifying selection acting on these genes most likely reflects an absence of specific parasites in Dro-sophila populations ½as opposed to termite colonies with higher pathogen pressures (Bulmerand Crozier 2004). Other instances of purifying selection have been observed for genes that are involved in binding to parasite polysaccharides, which are expected to evolve slower as compared to genes involved in binding to parasite proteins (Jigginsand Hurst2003; Littleet al. 2004; Jigginsand Kim2006).
We have previously demonstrated evidence for selec-tion on ArabidopsisRgenes involved in the specific rec-ognition of pathogen isolates (Bakker et al. 2006). These R genes tend to exhibit evidence of balancing selection, maintaining high numbers of alternative alleles that exhibit high levels of recombination but that are not unusually old. We additionally find evidence for selective sweeps, or partial selective sweeps, at a fewR gene loci. These results lead to a model in which selection maintains variation for some period of time, but not for an unusually long period of time, consistent with the trench-warfare model of plant-pathogen co-evolution (Stahl et al. 1999). R genes are at the top of the defense response pathway of Arabidopsis. We currently know nothing about the molecular evolution of other defense response genes in Arabidopsis.
Tests to detect selection assume selective neutrality as a null hypothesis, and the population is often assumed to be at a demographic equilibrium. Departures from such models are interpreted as indications of natural selection (Kreitman 2000). But systematic surveys of polymorphism across the Arabidopsis thaliana genome have shown that genomewide polymorphism data do not fit standard neutral models, even when nonequilib-rium demographic conditions are imposed (Nordborg
et al. 2005; Schmid et al. 2005). In the absence of a defensible null neutral model against which to test for selection, several authors have begun substituting an empirical distribution of genomewide polymorphism for a theoretical distribution as a baseline against which to identify unusual features of polymorphism in genes of interest, as previously advocated by Kreitman(2000) and by Nordborget al.(2005).
This ‘‘empirical’’ approach was employed by Bakker
et al.(2006) to investigate evolutionary scenarios for a set of 27 nucleotide binding site–leucine-rich repeat (NBS– LRR)Rgenes, as described above. An attempt was made to distinguish between three evolutionary scenarios: (1) long-term balancing selection, where a gene may have highly diverged alleles at intermediate frequencies and a high level of silent polymorphism; (2) positive selection on a favorable allele, which can generate a locus with few, relatively young alleles with extended haplotypes (some-times referred to as a selective sweep); and (3) purifying selection or functional constraint, which leads to low levels of nonsynonymous polymorphism and a commen-surately low rate of divergence between species. An empirical modeling approach was additionally used by Akeyet al. (2004) to identify eight disease genes that appear to be under selection in humans. They used polymorphism patterns obtained for 132 genes to dis-tinguish between the effects of demographic history and selection. Using an empirical approach without models, Fstvalues of individual SNPs were contrasted with the
empirical genomewide distribution of Fst; Akey et al.
(2002) identified 174 candidate genes as targets of selection. However, this study was limited by a relatively small sample size, representing only a few populations. In contrast, the analysis presented in this article for defense response genes inA. thalianarelies on empirical distributions obtained from genomewide polymor-phism patterns based on a worldwide sampling of this species.
lower numbers of protein variants and are in general less polymorphic than random genomic fragments, showing an overall pattern of purifying selection with a small number of instances of positive selection and adaptively driven substitutions.
MATERIALS AND METHODS
Sampling and data generation: DNA was extracted from
fresh leaves of 96A. thalianaaccessions (supplemental Table 1). The accessions were chosen to include a worldwide sample, where 50 were sampled from local populations in a hierarchi-cal fashion. The same set of 96 accessions has been extensively characterized for polymorphism in 876 genomic fragments that average 583 bp in length and represent 0.48 Mbp of the Arabidopsis genome (Nordborget al.2005). We selected 27
genes that are involved in the Arabidopsis defense response pathway (for a functional description of these genes, see sup-plemental Table 4). Defense response gene sequences for the 96A. thalianaaccessions were obtained by directly sequencing PCR amplification products. The 876 genomic fragments were selected so that they would cover the Arabidopsis genome with an average spacing of100 kb. Since the 27 defense response genes were not selected in this way, their distances to neigh-boring fragments are in general shorter than the distances between each of the 876 fragments and their nearest neighbors.
Primers were designed to amplify fragments of600 bp for 27 defense response genes using theA. thalianareference se-quence. We were able to obtain sequence information for most of the accessions for 27 defense response gene loci (supple-mental Tables 2 and 3). For 2 defense response genes more than 1 fragment was amplified: at1g66290 (ETR1; 3 fragments) and at4g26120 (NPR2; 2 fragments), which were subsequently concatenated. Studied regions predominantly consisted of exons, where the total length of coding sequences ranged between 159 and 1182 bp (supplemental Table 3). Twenty-two fragments contained 1 or more (parts of) introns and 3 fragments also contained noncoding sequences located out-side of the studied gene (supplemental Table 3). Sequences were generated on ABI 3700 automated sequencers (Applied Biosystems, Foster City, CA) by Genaissance Pharmaceuticals. All fragments were sequenced in both directions. Heterozy-gous sites were treated as missing data. Sequences have been submitted to the GenBank database (accession nos. EU404192–EU406770).
Data analysis:We compared the set of 27 defense response
genes with the set of 876 random fragments, for a number of population genetic summary statistics in a similar fashion to that done by Bakkeret al.(2006). A number of summary
sta-tistics were calculated using C11programs written by EGB: nucleotide diversityp(Neiand Li1979), number of
segregat-ing sites standardized by sequence lengthS, Tajima’sD(Tajima
1989),FST(Nei1973), and minimum number of
recombina-tions,Rh(Myersand Griffiths2003). Summary statistics
in-volving coding regions of defense response genes and random genomic fragments (i.e., 236 random genomic fragments with
.400 bp exon sequence) included numbers of synonymous (Ks) and nonsynymous (Ka) substitutions (Neiand Gojobori
1986), maximum number of synonymous substitutions be-tween accession pairs (Ksmax), and the number of protein
variants on the basis of nonsynonymous substitutions. We also calculatedp,S, and Tajima’sDfor synonymous and nonsyn-onymous sites separately. Confidence intervals for the esti-mates ofKa,Ks, andKa/Ksratio were obtained through Monte
Carlo simulation using the program K-estimator (Comeron
1999). Distributions of other above-mentioned summary statistics calculated for the set of defense response genes were compared with corresponding distributions obtained for the random fragments by doing Mann–WhitneyU-tests in R 1.8.1 (http://www.r-project.org/). In addition, we checked whether summary statistic values for the set of defense response genes would reside in the upper or lower 5th percentile of the corresponding distribution of statistics for the random frag-ments, as determined from the average values of 27 randomly sampled fragments, repeated 1000 times. While the 5% lower and upper tails together represent 10%, which would not be considered conservative, we believe that these tails do provide conservative cutoff values since the set of 876 reference fragments—and especially the set of 236 coding fragments— contains fragments that are under selection or are in close linkage to sites that are under selection. Analysis of this set of 876 reference fragments has shown fat tails for the distribu-tions of Tajima’sDand level of polymorphism ˆus, confirming
that a substantial fraction of fragments is under selection or is closely linked to a fragment under selection (Nordborget al.
2005).
Evidence for a recent selective sweep can be viewed as an extended region with reduced recombination surrounding the target of selection resulting in reduced variation at neighboring loci (Maynard-Smithand Haigh1974; Aguade´
et al.1989). We investigated defense response genes with low nucleotide diversity levels further for indications of a recent selective sweep according to Bakkeret al.(2006). This method
involves comparison of differences in silent nucleotide di-versity (ps) between these defense response genes and
frag-ments within 400 kb (obtained from http://walnut.usc.edu/ 2010/), followed by a similar analysis in the 59and 39directions topsdifferences for each of the 876 random fragments and its
nearest neighboring fragments on both sides.
We used a second approach to identify partial selective sweep candidates. This approach involved a genomewide scan of haplotype sharing according to Toomajianet al.(2006). We
used for this purpose polymorphism data from 1204 fragments that come from an ongoing survey of genomic DNA sequence andRgene polymorphism (Nordborget al.2005). The
den-sity of these sequenced regions along each chromosome is sufficient to identify unusually long identical haplotypes among subsets of these individuals. The haplotype-sharing ap-proach applied here is based on the assumption that chromo-somes that are identical by descent (IBD) at a polymorphic site must also share a short region surrounding that site. There-fore, the age of the shared allele at the polymorphic site and the recombination rate in the region influence the length of this IBD region (Nordborg and Tavare´ 2002; Innan and
Nordborg2003). We compared haplotype sharing around
each defense response gene with what is observed in the rest of the genome in a similar fashion as Bakkeret al.(2006). See
Toomajianet al.(2006) for a more detailed description of the
haplotype sharing statistic.
We used orthologous sequences from the 43A. lyrata as-sembly (Arabidopsis lyrata Sequencing Consortium,
unpublished data) as outgroup sequences for Ka/Ks and
McDonald–Kreitman tests (McDonaldand Kreitman1991).
used a publishedA. lyratasequence (GenBank accession no. AF263721; Hauseret al.2001) as an outgroup sequence.
RESULTS
Overview of polymorphism patterns in defense re-sponse genes: We obtained sequence information for 27 defense response genes, most of which are involved in the SA-, JA-, or ET-dependent signal transduction pathways or are part of downstream pathways that can be triggered after gene-for-gene interaction (Figure 1; sup-plemental Table 4). The genes vary widely in their func-tions in these pathways but can be subdivided into five categories on the basis of their annotation: (1) transcrip-tion regulatranscrip-tion, (2)Rgene response, (3) phytoalexin/ PR protein production, (4) cell-death regulation, and (5) function in intermediate stage of pathway (Table 1). Since each category is involved in a different stage of these pathways, they are expected to experience ent selective pressures. However, no significant
differ-ences were observed among these groups on the basis of the averages of summary statistics after Bonferroni correction (P .0.05; Ksmax,p,S, Tajima’sD,Rh, Fst,
number of protein variants, and interspecies,Ka/Ks). The complete group of 27 genes displays a wide range of evolutionary histories. A graphical overview of de-fense response gene evolution was obtained by plotting neighbor-joining trees on the basis of silent sites (Nei and Gojobori1986) using MEGA version 2.1 (Kumar
et al. 2001; Figure 2). This figure reveals substantial variation in tree shape and depth. Comparison with the allele (haplotype) frequency distribution of the set of 876 random genomic fragments (Figure 3) shows that defense response genes have a significantly higher minor allele frequency for synonymous SNPs (Mann–Whitney U-test, P , 0.05) and a marginally lower minor allele frequency for nonsynonymous SNPs (Mann–Whitney U-test,P¼0.054). These patterns reflect an excess of low frequency (minor allele frequency, P ,0.05) nonsyn-onymous SNPs compared to synnonsyn-onymous SNPs (62% vs. 43%, respectively) for the defense response genes, whereas for the reference loci this difference is much smaller (58%vs.52%). However, both the reference loci and the defense response genes appear to have a sig-nificant excess of low frequency nonsynonymous alleles compared to synonymous alleles (Mann–WhitneyU-test, P , 0.0001 and P , 0.0005, respectively), indicating purifying selection acting on nonsynonymous mutations. Principal component analysis (PCA) based on seven summary statistics (Ksmax—the maximum number of synonymous substitutions between accessions pairs—p, S, Tajima’sD,Rh,Fst, and number of protein variants)
shows that the 27 defense response genes do not fall into distinct clusters but cover a range of possible evolution-ary states (Figure 4). The first principal component (PC1, 44% of total variation) differentiates the set of defense response genes on the basis of Ksmax (maxi-mum interallelic synonymous divergence, a measure of the maximum depth of a gene tree), Tajima’sD(test for selection), p(nucleotide diversity),S (number of seg-regating sites) and Rh (recombination frequency) in such a way that defense response genes with high PC1 values are characterized by alleles that are highly differentiated one from another at the sequence level (Table 2). Small PC1 values, in contrast, indicate low levels of polymorphism. The second principal compo-nent (PC2, 20% of total variation) differentiates the set of defense response genes on the basis of Tajima’sD(test for selection), Fst (population differentiation), and
number of protein variants so that defense response genes with high PC2 values are characterized by a rela-tive excess of alleles at intermediate frequency and high population differentiation (Table 2). Small PC2 values can indicate highly skewed allele frequency distributions and interpopulational uniformity. Candidates for bal-ancing selection and selective sweep on the basis of analyses of population genetic summary statistics (Table
Figure1.—A model showing putative positions of 17
de-fense genes in the salicylate (SA)-, jasmonate ( JA)-, and eth-ylene (E)-dependent signaling pathways inA. thaliana. PAD4, EDS1, EDS5,NPR1,BGL2, EDR1,CPR5,ACD1,ACD2, DND1, andCOI1 are shown as in Glazebrook(2001) and G
laze-brooket al. (2003). PAL1 (Mauch-Mani and Slusarenko
1996) andFAD3 (McConnand Browse1996) are involved
in the synthesis of salicylic acid and jasmonic acid, respec-tively. ETR1is a histidine kinase that binds ethylene (Guo
and Ecker2004) and negatively regulates theEIN3
transcrip-tion factor (Gagneet al.2004).AHBP1is a transcription
fac-tor that interacts withNPR1(Fanand Dong2002; Pieterse
and VanLoon2004).BGL1is a beta-glucosidase enzyme
up-regulated by jasmonic acid, but not by salicylic acid (Stotz
et al. 2000). The following 10 defense genes do not currently fit clearly in the model:NPR2 andNPR3are ankyrin repeat proteins similar to NPR1 in predicted function (Liu et al.
2005).NHL1andNHL3are plasma membrane proteins with probable roles in the hypersensitive response (Varetet al.
2003; Zhenget al.2004).PBS2(Holtet al.2005) andPBS1
(Shao et al. 2003) interact directly with specific R genes.
PAD3is a cytochrome P450 required for camalexin synthesis (Zhouet al. 1999).ATR1andESPare enzymes involved in
glu-cosinolate synthesis (Nauret al.2003) and hydrolysis (L am-brixet al. 2001), respectively. GL1 is a transcription factor
1) are indicated with different symbols in the PCA plot (Figure 4).
Pseudogenes and nonfunctional alleles: Alleles
rendered nonfunctional due to mutations causing frameshifts and/or premature stop codons were ob-served for five defense response loci, whereEDS5stood out with 36 (38%) plant accessions with nonfunctional alleles. Nonfunctional alleles were a result of a stop codon mutation, coinciding with a major clade in the gene tree (Figure 2). Four loci (ESP,ETR1,EDS1, and PAD4) had nonfunctional alleles in 1 or 2 of the 96 plant accessions. The number of loci with nonfunctional alleles did not differ significantly between the set of defense response genes and the set of reference loci (18.5%vs.24.0%; Mann–WhitneyU-test,P¼0.52).
Population structure:Genomewide patterns of poly-morphism that depart from neutral expectations are generally assumed to reflect demographic history. In contrast, loci that exhibit departures from neutrality but have atypical geographic structure may be candidates for selection. To investigate this possibility, we measured population differentiation (Fst) for the defense
re-sponse genes across eight groups of accessions, which were previously identified by Structure (Falush et al. 2003) as distinguishable population clusters (N ord-borg et al. 2005; supplemental Table 1). These eight groups followed plausible geographic boundaries. Fst
values were calculated using alleles identified on the
Figure 2.—Neighbor-joining
trees based on silent sites (Nei
and Gojobori 1986), sorted on
the basis of the level of allelic diver-gence. 1, at1g52400 (BGL1); 2, at5g13160 (PBS1); 3, at3g52430 (PAD4); 4, at3g57260 (BGL2); 5, at3g20770 (EIN3); 6, at2g39940 (COI1); 7, at5g51700 (PBS2); 8, at5g06950 (AHBP1); 9, at4g31500 (CYP83B1); 10, at4g39030 (EDS5); 11, at5g64930 (CPR5); 12, at3g11660 (NHL1); 13, at1g54040 (ESP); 14, at3g26830 (PAD3); 15, at3g48090 (EDS1); 16, at2g29980 (FAD3); 17, at5g06320 (NHL3); 18, at1g08720 (EDR1); 19, at5g45110 (NPR3); 20, at3g44880 (ACD1); 21, at4g37000 (ACD2); 22, at4g26120 (NPR2); 23, at3g27920 (GL1); 24, at1g64280 (NPR1); 25, at2g37040 (PAL1); 26, at5g15410 (DND1); and 27, at1g66290 (ETR1). Scale bar reflects 0.01 substitutions per site.
Figure 3.—Allele frequency distribution for synonymous
and nonsynonymous SNPs observed for the set of 27 defense response genes and the set of 876 random genomic fragments.
Figure4.—Principal component analysis for 27 defense
re-sponse genes based on the correlation matrix of seven sum-mary statistics (Ksmax,p, S, Tajima’s D, Rh, Fst, number of
basis of both nucleotide substitutions and indels. No significant difference was observed between defense response genes and the genomewideFst distributions
(P ¼0.49). Only one defense response gene (ACD1) had aFstvalue below the lower 5th percentile of the
empirical distribution, and none hadFst values in the
upper 5th percentile of the empirical distribution. These results argue strongly against widespread selection corre-sponding to eight geographical regions (see Nordborg
et al.2005).
Comparisons with empirical distributions:Unusually large values of the summary statisticsKsmax (allelic diver-gence),p(nucleotide diversity),S(number of segregat-ing sites), Tajima’s D (test for selection), number of protein variants, and interallelicKa/Ksratios (compar-ison of number of amino acid replacement changes with silent changes) are possible indicators of a locus being under the influence of balancing selection, whereas un-usually small values for these summary statistics suggest either a recent selective sweep or a purifying selection. Comparison of the 27 defense response genes with the reference loci revealed that defense response loci have fewer protein variants (mean¼2.78) than the reference loci (mean¼5.38; Mann–WhitneyU-test,P,0.0005), and lower nonsynonymous site diversity (pnonsyn ¼ 0.0008 for defense response genesvs.pnonsyn¼0.0017 for the reference loci,P,0.005) and lower number of nonsynonymous segregating sites (mean¼0.0071 for defense response genes vs. mean ¼ 0.0132 for the reference loci; Mann–WhitneyU-test,P,0.05).
Reduction in the number of protein alleles by a recent selective sweep can be distinguished from purifying selection against amino acid replacement mutations by asking whether synonymous nucleotide diversity is also reduced. Synonymous site diversity is slightly but not significantly higher for the defense response genes as compared to the reference loci (psyn¼0.0020vs. 0.0017, respectively; P ¼0.27). Thus, the polymorphism data favor purifying selection.
The defense response genes and the reference loci did not differ significantly for their Tajima’sD (synon-ymous and nonsynon(synon-ymous polymorphism combined):
0.5065 and 0.8338 (P ¼ 0.07) for the defense re-sponse genesvs.the reference loci, respectively. Defense response genes are both in the lower and upper 5% tail of the empirical distribution, but the variance of the Tajima’sDestimates for the individual defense loci does not differ from that of the empirical distribution (F-test, P ¼0.47). For ETR1 we found that one of the three concatenated fragments (starting at position 1.24739369) had a significantly high Tajima’s D value (2.2573), in the upper 2.5th percentile of the empirical distribution. Further analysis revealed no significant difference be-tween the set of 27 defense response genes and the empirical distribution for Tajima’s Dvalues calculated on the basis of nonsynonymous sites only (P ¼0.23). Average Tajima’s D calculated on the basis of synony-mous sites, however, is more positive (0.2028) than the empirical distribution (0.4394;P¼0.41). Even though this difference is not significant, this positive shift in Tajima’sDis a general tendency of the group of 27 genes rather than the presence of a larger than expected number of positive outliers, since we observed only 3 defense response genes (ETR1,PAD3, andDND1) with a synonymous Tajima’sDvalue in the upper 5% tail of the corresponding empirical distribution (P¼0.31).
We did not observe a significant difference inRh, the minimum number of recombination events per base pair, between the defense response genes and the em-pirical distribution (Mann–Whitney U-test, P ¼0.19). We observed anRhvalue of zero for 20 defense response genes (74.1%), but this percentage is not significantly different than for the empirical distribution (68.7%; Mann–Whitney U-test,P ¼0.55). This result stands in strong contrast with our similar analysis of R genes, where recombination rates were significantly higher as compared to the empirical distribution for a given
p-value (Bakkeret al.2006).
Ksmax, a measure of the maximum number of syn-onymous substitutions between any pair of accessions at a locus, provides a gauge of the time to the common ancestry of the sample. Large values indicate the pres-ence of long-lived alleles at a locus, whereas small values imply a recent common ancestry of alleles. Compared to the empirical distribution, the most diverged alleles at a defense response locus do not tend to be unusually old (P ¼ 0.44). For 1 defense response gene (BGL1) we observed Ksmax¼0, but this frequency of zero values (3.7%) in the set of 27 defense response genes is not significantly different from the empirical distribution (8.8%; Mann–Whitney U-test, P ¼ 0.39). We did not observe interallelicKa/Ksvalues significantly.1 for any defense response locus, which if present might indicate adaptive divergence of alleles (data not shown).
Comparisons with outgroup sequences:We have ob-tained corresponding A. lyrataorthologous sequences for all 27 defense response genes.Ka/Ksvalues between
A. thaliana and A. lyrata ranged from zero for PAL1 (At2g37040) to 1.21 forPAD3(At3g26830). None of the TABLE 2
Loadings for the summary statistics that constituted the first and second principal component in a principal component
analysis of the set of 27 defense response genes
Summary statistic PC1 PC2
Ksmax 0.389 0.103
p 0.549 0.115
S 0.482 0.258
Tajima’sD 0.362 0.435
Rh 0.399 0.025
Fst 0.140 0.631
27 defense response genes exhibited aKa/Ksvalue sig-nificantly .1, indicating an absence of recurrent adaptive substitution. This observation was confirmed by nonsignificant McDonald–Kreitman test results for all 27 tested defense response loci, indicating no significant deviations from the neutral expectation. This means that the majority of substitutions are neutral or nearly neutral and that the most prevalent type of selection acting on these genes is purifying selection.
Search for positive selection in defense response genes:Even though the distributions of summary statis-tics for the 27 defense response genes conform reason-ably well to the empirical distribution, 8 loci can be identified as having a combination of values that might qualify them as candidates for selective sweeps.PAL1, COI1,EIN3,PAD4,BGL2,CYP83B1,PBS1, andCPR5are all observed to have lowpsyn,Rh,Ksmax, Tajima’sD, and number of protein variants. For 3 of these genes (COI1, PAD4, andCYP83B1), nucleotide diversity values were located in the lower 5% tail of the empirical distribu-tion. None of these 8 defense response genes exhibit evidence for possessing recombinant alleles (Rh¼0), but with low levels of polymorphism the ability to detect recombination is also low. Three of these 8 defense response loci (COI1,CYP83B1, andPBS1) code for only one protein variant, whereas the remaining 5 code for 2–4 protein variants.
To further investigate possible selective sweeps for these loci, we took advantage of the reference loci data-base to ask whether polymorphisms in loci flanking these genes are similarly low in their levels of poly-morphism. Silent nucleotide diversity (ps) was calcu-lated for 5–19 flanking fragments located at distances within 400 kb on each side of these eight defense re-sponse genes and the differences inpsbetween defense response genes and their flanking fragments were plotted against physical distance separating the loci (Figure 5). A similar analysis was carried out for the set of 876 random fragments. In these cases, we analyzedps differences between each fragment and its two nearest neighbors at distances within 400 kb. For the set of 876 random fragments,psdifferences did not increase sig-nificantly at increasing physical distance (Figure 5).ps differences between the eight defense response genes and their flanking fragments were concentrated below the median of the empirical distribution: 51 fragments below the medianvs.16 fragments above (chi-square test of independence, P , 0.0001; Figure 5), which is an indication that these loci are imbedded in larger regions of low polymorphism.
For none of the eight defense response genes did we consistently observepsdifferences below the lower 5% threshold on the basis of observations for the 876 ran-dom fragments. However, forPAD4,PAL1, andPBS1we observed lowpsdifferences with all fragments on both sides (within 170 kb) below the median of the random fragments distribution. A similar, but less obvious
pat-tern was observed for the other defense response genes, wherepsdifferences with fragments flanking one side of the defense response gene are in general low (below the median of the empirical distribution), whereasps differ-ences are relatively high on the other side. For one defense response gene,EIN3, we observed largeps dif-ferences with fragments on both sides. Such a pattern might be expected at a locus following a selective sweep if the recombination rate is sufficiently high to overcome linkage disequilibrium induced by selection at neigh-boring loci.
We also looked for evidence of partial selective sweeps of individual alleles using a haplotype-sharing approach (Toomajianet al.2006; also seematerials and meth-odssection). A haplotype sharing score was assigned to
each of 265 defense response alleles where the minor allele frequency at the corresponding polymorphic site was $2 (Figure 6). A score was also assigned to each allele selected on the basis of the same criterion from the 1204 reference fragments. A sliding window along al-leles ordered by frequency was used to produce the 95th percentile of the random fragments distribution. Al-though only nine (3% of 265) defense response gene
Figure 5.—Difference in silent nucleotide diversity
alleles (from five defense response genes) between the frequency of 0.05 and 0.95 are located above this percentile, we consider these candidates for recent se-lection because the tail of the empirical distribution may itself contain adaptive alleles. One allele from each of two different genes (AHBP1 and DND1) is associated with long haplotypes. There are multiple (2–3) alleles associated with long haplotypes for three other genes (ESP,EDS5, andNHL3). The two outlying alleles from ESP are nonoverlapping groups of accessions corre-sponding with different geographical regions, whereas the two and three outlying alleles fromEDS5andNHL3, respectively, share the same groups of accessions. Six of these alleles have a haplotype sharing score in the top 2.5% of the empirical distribution (AHBP1, one allele of ESP, one allele ofEDS5, and all three alleles ofNHL3). For only two loci is the allele at a relatively low frequency (,8 accessions;ESPandEDS5), the rest of the alleles are at frequencies varying between 23 and 88 accessions.
The defense response genes have fewer protein var-iants and are less variable than random genomic frag-ments, indicating either a purifying selection or a recent selective sweep. However, for a few of these genes (marked with an open triangle in the PCA graph, Figure 3) we observed high positive Tajima’sDandKsmax val-ues for synonymous sites. The best candidate for bal-ancing selection isETR1. The allelic divergence (Ksmax) of this locus is the largest of all the 27 defense response genes. In addition, one of the three concatenated frag-ments of this locus has a significant Tajima’s D value (2.2573). The other candidates for balancing selection areNPR1,PAD3,NHL3, andDND1, since their Tajima’sD values for synonymous sites are located in the upper 5% tail of the corresponding empirical distribution. With
the exception ofPAD3(Ksmax¼0.0255), all these genes have intermediate to largeKsmax values (0.0529–0.0929). Two other genes,ACD1andDND1, have relatively high Ksmax values (.0.09). One of the diverged alleles of
ACD1 occurs at a low frequency (as indicated by a low Tajima’s Dfor the locus), making it a less likely candi-date for selection.
DISCUSSION
Overview of polymorphism patterns: Defense
re-sponse genes tend to maintain lower levels of diversity as compared to the reference loci. These lower levels of diversity most likely reflect purifying selection acting on most defense response genes (see next section). This is in sharp contrast toRgenes, which in general tend to maintain high levels of diversity over short periods of time (Bakker et al.2006). Although defense response genes are not a homogeneous set with respect to function as is the case forRgenes, it is remarkable that as a group they all show low levels of diversity, indicative of purifying selection. For a few defense response loci we found evidence for (partial) selective sweeps and/or transient balancing selection. However, in general, de-fense response genes appear to be experiencing purify-ing selection and do not appear to differ from any randomly sampled genomic region with respect to their evolutionary scenario.
Defense response genes are under purifying selec-tion: Many of the defense response genes in our study have specific roles in signal transduction pathways and it is therefore expected that at least some of them are experiencing selective constraint. Purifying selection is characterized by relatively low nonsynonymous nucleo-tide diversity and low numbers of protein variants. Be-cause there is generally much less selective constraint on synonymous changes, synonymous nucleotide diversity values should exceed nonsynonymous nucleotide di-versity. The defense response genes as a group display lower levels of nonsynonymous nucleotide diversity and have fewer nonsynonymous segregating sites compared with the reference loci. Synonymous sites, in contrast, do not differ significantly between the sets of loci for nucleotide diversity or segregating sites, indicating that the low nonsynonymous polymorphism levels are likely due to functional constraint. Comparison withA. lyrata orthologous sequences provided further evidence for purifying selection: for none of the 27 defense response genes did we observe a Ka/Ks value significantly .1 and/or a significant McDonald–Kreitman test result.
Defense response genes under strong purifying selec-tion arePBS1,FAD3,CYP83B1,COI1,PBS2, andDND1, which code for only one protein variant.PBS1codes for a protein kinase, which is involved in the detection of avrgenes andRgenes response (Shaoet al. 2003).FAD3 codes for an omega-3 fatty acid desaturase, which syn-thesizes linolenic acid, a JA precursor (McConn and
Figure6.—Haplotype sharing for 12,038 alleles from a set
Browse 1996).CYP83B1is involved in indole glucosi-nolate biosynthesis (Naur et al. 2003) and COI1 is involved in proteolysis (Xieet al. 1998).PBS2is a resis-tance signaling protein that acts uponRgenes response (Holtet al. 2005). We suggest that these six are likely to have histories dominated by purifying selection, on the basis of their relatively high levels of synonymous nu-cleotide diversity. One of these loci (DND1), however, has a relatively large positive Tajima’s Dvalue, raising the possibility of a selectively maintained polymorphism in another region of the gene.
It needs mention that several defense response genes also have other housekeeping functions, which could explain the conserved nature of these genes. However, of the six defense response genes that appear to be ex-periencing purifying selection, there is only one gene (CYP83B1) that is also involved in processes other than defense response (supplemental Table 4). It is therefore possible that this gene is under purifying selection because of its other essential functions.
Defense response genes do not experience balancing selection, selective sweeps:Previous evolutionary anal-yses of a similar set ofRgenes in Arabidopsis revealed indications for transient balancing selection and partial selective sweeps (Bakkeret al.2006). Our data show that there is not much evidence for selective sweeps and balancing selection in the set of 27 defense response genes. However, for a proper comparison with the pre-viously published set of 27Rgenes (Bakkeret al.2006), we also have to look at these more complicated signs of selection. On the basis of comparisons with the refer-ence loci, we conclude that as a group, purifying selec-tion appears to be the dominant form of selecselec-tion. For a small number of defense response genes, however, we have indications for positive selection (selective sweep) and balancing selection.
Eight defense response genes have a combination of lowpsyn,Rh,Ksmax, Tajima’sD, and number of protein variants (PAL1, COI1, EIN3, PAD4, BGL2, CYP83B1, PBS1, andCPR5), which together may indicate recent selective sweeps. Although the nucleotide diversity at these loci was lower than at other apparently neutrally evolving A. thaliana functional genes (e.g., Kuittinen and Aguade´2000; Aguade´2001; Hauseret al.2001), it was not significantly different from the reference loci. We therefore believe that these defense response genes have undergone purifying selection. We identified only one candidate for a recent selective sweep (PAD4) on the basis of shared low diversity levels with neighboring genes. This gene is involved in activation of SA accu-mulation, camalexin synthesis, PR gene expression, and Rgenes response (Zhouet al.1998). We investigated the possibility of partial selective sweeps by using a haplo-type sharing approach (Toomajian et al. 2006). This analysis revealed several possible instances of a partial selective sweep: AHBP1 (transcription factor activity; Fan and Dong 2002; Pieterseand Van Loon2004)
with a haplotype near fixation (92% of accessions) and DND1(cell-death regulator; Cloughet al. 2000) with an extended haplotype at intermediate frequency (60% of accessions), corresponding with a major clade in the gene tree. This latter locus has relatively highKsmax and (synonymous) Tajima’sDvalues and therefore appears to be under balancing selection or is closely linked to a site under balancing selection. The fact that we observe evidence for a partial selective sweep and for balancing selection is reminiscent of observations we previously made for theRgeneRPM1, where there is evidence for historical allele frequency fluctuations (Stahl et al. 1999) and for the R genes at1g56540 and at5g63020 (Bakkeret al. 2006). It is possible that this defense re-sponse gene has experienced a recent rise in the fre-quency of a favorable allele resulting in an extended haplotype sharing surrounding the site(s) under selec-tion. For three defense response genes:ESP(promotes the hydrolysis of glucosinolates to nitriles; Lambrixet al. 2001), EDS5 (transport of intermediates for SA bio-synthesis,Rgenes response; Nawrathet al.2002), and
NHL3(transport of intermediates for SA biosynthesis,R genes response; Nawrathet al.2002), we observed.1 extended haplotype. Although multiple extended hap-lotypes could be the result of multiple partial sweeps at the same gene, they may simply result from the complex history of a single sweep. In fact, on the basis of their observations regarding a clustering of extreme values of a similar haplotype sharing statistic, Voightet al.(2006) have suggested that the clustering of extreme alleles may be a strong sign that at least one of the alleles is a true positive.
For EDS5 we observed 38% nonfunctional alleles coinciding with a major clade in the gene tree, whereas four other loci (ESP, ETR1, EDS1, and PAD4) had nonfunctional alleles in 1 or 2 of the 96 plant accessions. Our R gene study detected nonfunctional alleles in 17 of the 27 studied R genes in a range from 1.15 to 32.8% (Bakker et al. 2006). Nonfunctional alleles occurring at high abundances were observed to either coincide with one or more major clades (7Rgenes) or to be scattered over the entire gene tree (2R genes). Highly differentiated major clades with nonfunctional alleles in R genes are expected to be maintained by balancing selection due to a cost of resistance (Stahl
et al. 1999; Tianet al. 2002, 2003). Possibly a similar mechanism has resulted in the maintenance of a clade with nonfunctional alleles inEDS5.
Four defense response genes appear to be under tran-sient balancing selection:ETR1(ethylene-binding activity; Guoand Ecker2004),NPR1(transcription regulation; Zhanget al. 1999),DND1(cell-death regulation; Clough
summary statistics. Only synonymous Tajima’s Dvalues of these defense response genes showed a significant deviation from the empirical distribution.
Comparisons between defense response genes andR
genes:Comparison of these 27 defense response genes with 27 previously studiedRgenes (Bakkeret al.2006) shows that R genes display in general a significantly higher nonsynonymous nucleotide diversity (0.007vs. 8.163 104;P ,0.0001), number of nonsynonymous segregating sites (0.028vs.0.007;P ,0.0001), recom-bination frequency (0.0037vs.0.0009;P,0.005), and number of protein variants (14.2vs. 2.8;P ,0.0001). Clearly,R genes maintain higher levels of protein var-iation due to natural selection acting to maintain non-synonymous mutations and an elevated recombination frequency. It is believed that transient balancing selec-tion is the main selective force acting onRgenes causing high levels of protein variation maintained over in-termediate periods of time (Bakkeret al. 2006). This observation was confirmed by the resequencing study of 20 Arabidopsis accessions of Clarket al. (2007), which identified strong signatures of selection on NBS–LRR typeRgenes. It is widely argued thatRgenes experience this type of selection because of their direct interaction with pathogen avirulence ligands. In contrast, defense response genes, which do not interact with pathogen avirulence ligands, are substantially less variable. How-ever, the near complete absence of positive and/or balancing selection signatures in this set of defense response genes is unexpected since evolutionary analy-ses of genes involved in other (defense response) path-ways have shown varying evolutionary scenarios for groups of genes involved in those pathways. For exam-ple, studies of genes involved in the anthocyanin bio-synthetic pathway have shown differences in selective constraint between upstream and downstream genes in the pathway and differences in evolutionary rates between regulatory and structural genes in maize, snapdragon, and morning glory (Purugganan and Wessler 1994; Rausheret al.1999). In their study of genes involved in the floral developmental pathway in Arabidopsis, Olsenet al.(2002) observed that patterns of molecular evolution differ among regulatory genes with earlier acting genes exhibiting evidence of adaptive evolution. Similar patterns of evolution have been observed for various defense response pathways both in plants and in animals. Studies of defense response genes involved in the innate immunity response of some insect species have shown evidence for adaptive evolu-tion on genes downstream in this defense response pathway (Schlenke and Begun 2003; Bulmer and Crozier2004). Several genes involved in the glucosi-nolate–myrosinase system show high levels of adaptive variation and evidence for balancing selection (reviewed by Kliebensteinet al. 2005). However, in our study of 27 defense response genes inA. thalianawe did not observe any significant difference in population genetic
sum-mary statistics between genes involved in different parts of SA, JA, and ET signaling pathways. A possible explanation for this observation is that Arabidopsis SA, JA, and ET signaling pathways do not produce as high a number of variable end products as the glucosinolate– myrosinase system. A. thaliana contains .30 different glucosinolate structures (Kliebenstein et al. 2001; Reichelt et al. 2002), which are the result of genetic variation in glucosinolate core pathway and regulatory genes (Kliebenstein et al. 2002). It should be noted, however, that there is an overlap between the glucosi-nolate–myrosinase system and the SA and JA signaling pathways: jasmonates increase total glucosinolate con-centration whereas salicylate stimulates glucosinolate accumulation (Kliebensteinet al.2002). Two of the 27 defense response genes from our study are also de-scribed as being involved in the Arabidopsis glucosino-late biosynthesis (i.e.,CYP83B1andESP; Kliebenstein
et al.2005). These 2 genes have not been characterized as candidates for balancing selection in previous analy-ses of the glucosinolate–biosynthesis system (K lieben-steinet al. 2005), confirming our observations of low levels of variation observed for these genes. Neverthe-less, it is expected that genes upstream in the SA, JA, and ET signaling pathways experience less directional selec-tion as compared to downstream and regulatory genes. A more detailed analysis of the Clarket al. (2007) data encompassing all defense response genes in combina-tion with our data of 27 defense response genes might be able to reveal such a pattern for defense response genes. Another possible explanation for the observed differ-ences in evolutionary histories between defense re-sponse genes and Rgenes is that there is a difference in gene structure between the two groups of genes that could have an effect on to what extent high levels of di-versity can be maintained. The LRR region in Rgenes consists of a varying number of leucine-rich repeat modules.Rgene specificity is believed to have evolved through LRR module indels and/or mutations affecting amino acid changes in the solvent-exposed parts of the
b-sheet structure ( Jonesand Jones1996). Because of the modular nature of the LRR region it is expected that more nonsynonymous mutations are tolerated in R genes as compared to defense response genes, which do not possess such a modular region.
of these genes. Additional polymorphisms from the unsequenced regions of these genes will likely provide additional insight into how selection acts on these genes. The wide variety of gene tree topologies, time-depths, and summary statistics for the 27 defense response genes may reflect a continuum of evolutionary histories, with contributions from both stochastic forces and selection (both positive and negative). Yet, except for greater pur-ifying selection on amino acid replacements, in almost every other respect this set of defense signaling pathway genes are indistinguishable from a large panel of ran-domly chosen loci spread across the entire Arabidopsis genome. This ‘‘unremarkable’’ pattern of polymorphism for this group of genes is surprising given that genes in other (defense response) pathways do seem to experi-ence different levels of selective pressure. Genes involved in the SA, JA, and ET signaling pathways in general ex-perience high levels of selective constraint possibly be-cause of the conserved nature of the end products of these pathways.
We thank Tina Hu and Keyan Zhao for their help with primer design, Matt Horton and Brian Mack for their help with editing of sequence data, and Arnar Palsson for statistical help. We thank the Department of Energy’s Joint Genome Institute for sequencing the
Arabidopsis lyratagenome and allowing us to use the 43draft assembly. This work was supported by a National Institutes of Health post-doctoral fellowship to C.T., a National Science Foundation grant (DEB-0115062), and a National Institutes of Health grant (GM-57994) to J.B. and M.K.
LITERATURE CITED
Aarts, N., M. Metz, E. Holub, B. J. Staskawicz, M. J. Danielset al.,
1998 Different requirements forEDS1andNDR1by disease
re-sistance genes define at least two R gene-mediated signaling
path-ways inArabidopsis.Proc. Natl. Acad. Sci. USA95:10306–10311.
Aguade´, M., 2001 Nucleotide sequence variation at two genes of the
phenylpropanoid pathway, theFAH1andF3Hgenes, in
Arabidop-sis thaliana. Mol. Biol. Evol.18:1–9.
Aguade´, M., N. Miyashitaand C. F. Langley, 1989 Reduced
var-iation in theyellow-achaete-scuteregion in natural populations of
Drosophila melanogaster.Genetics122:607–615.
Akey, J. M., G. Zhang, K. Zhang, L. Jin and M. D. Shriver,
2002 Interrogating a high-density SNP map for signatures of
natural selection. Genome Res.12:1805–1814.
Akey, J. M., M. A. Eberle, M. J. Rieder, C. S. Carlson, M. D. Shriver
et al., 2004 Population history and natural selection shape
pat-terns of genetic variation in 132 genes. PLoS Biol.2(10): e286.
Bakker, E. G., C. Toomajian, M. Kreitman and J. Bergelson,
2006 A genome-wide survey ofRgene polymorphisms in
Arabi-dopsis thaliana.Plant Cell18:1803–1818.
Bulmer, M. S., and R. H. Crozier, 2004 Duplication and
diversify-ing selection among termite antifungal peptides. Mol. Biol. Evol.
21:2256–2264.
Clark, R. M., G. Schweikert, C. Toomajian, S. Ossowski, G. Zeller
et al., 2007 Common sequence polymorphisms shaping genetic
diversity inArabidopsis thaliana.Science31:338–342.
Clough, S. J., K. A. Fengler, I.-C. Yu, B. Lippok, R. K. Smith, Jr.et al.,
2000 The Arabidopsisdnd1‘‘defense, no death’’ gene encodes
a mutated cyclic nucleotide-gated ion channel. Proc. Natl. Acad.
Sci. USA97:9323–9328.
Comeron, J. M., 1999 K-Estimator: calculation of the number of
nu-cleotide substitutions per site and the confidence intervals.
Bio-informatics15:763–764.
Dangl, J. L., and J. D. Jones, 2001 Plant pathogens and integrated
defense responses to infection. Nature411:826–833.
Durrant, W. E., and X. Dong, 2004 Systemic acquired resistance.
Annu. Rev. Phytopathol.42:185–209.
Falush, D., M. Stephensand J. K. Pritchard, 2003 Inference of
population structure using multilocus genotype data: linked loci
and correlated allele frequencies. Genetics164:1567–1587.
Fan, W., and X. Dong, 2002 In vivo interaction betweenNPR1and
transcription factorAHBP1leads to salicylic acid-mediated gene
activation in Arabidopsis. Plant Cell14:1377–1389.
Flor, H. H., 1956 The complimentary genic systems in flax and flax
rust. Adv. Genet.8:29–54.
Gagne, J. M., J. Smalle, D. J. Gingerich, J. M. Walker, S. Yooet al.,
2004 Arabidopsis EIN3-binding F-box 1 and 2 form ubiquitin-protein ligases that repress ethylene action and promote growth
by directingEIN3degradation. Proc. Natl. Acad. Sci. USA101:
6803–6808.
Glazebrook, J., 2001 Genes controlling expression of defense
re-sponses inArabidopsis—2001 status. Curr. Opin. Plant Biol. 4:
301–308.
Glazebrook, J., W. Chen, B. Estes, H. Chang, C. Nawrathet al.,
2003 Topology of the network integrating salicylate and
jasmo-nate signal transduction derived from global expression
pheno-typing. Plant J.34:217–228.
Guo, H., and J. R. Ecker, 2004 The ethylene signaling pathway: new
insights. Curr. Opin. Plant Biol.7:40–49.
Hauser, M.-T., B. Harrand C. Schlotterer, 2001 Trichome
distri-bution inArabidopsis thalianaand its close relativeArabidopsis
lyr-ata: molecular analysis of the candidate gene Glabrous1. Mol.
Biol. Evol.18:1754–1763.
Holt, III, B. F., Y. Belkhadirand J. L. Dangl, 2005 Antagonistic
control of disease resistance protein stability in the plant
im-mune system. Science309:929–932.
Innan, H., and M. Nordborg, 2003 The extent of linkage
disequi-librium and haplotype sharing around a polymorphic site.
Genet-ics165:437–444.
Jiggins, F. M., and G. D. Hurst, 2003 The evolution of parasite
rec-ognition genes in the innate immune system: purifying selection onDrosophila melanogasterpeptidoglycan recognition proteins. J.
Mol. Evol.57:598–605.
Jiggins, F., and K.-W. Kim, 2005 The evolution of antifungal
pepti-des in Drosophila.Genetics171:1847–1859.
Jiggins, F., and K.-W. Kim, 2006 Contrasting evolutionary patterns
in Drosophila immune receptors. J. Mol. Evol.63:769–780.
Jiggins, F., and F.-W. Kim, 2007 A screen for immunity genes
evolv-ing under positive selection in Drosophila.J. Evol. Biol.20:965–
970.
Jones, D. A., and J. D. G. Jones, 1996 The roles of leucine-rich
re-peat proteins in plant defences. Adv. Bot. Res. Adv. Plant Pathol.
24:89–167.
Kliebenstein, D. J., J. Kroymann, P. D. Brown, A. Figuth, D.
Petersenet al., 2001 Genetic control of natural variation in
Arabidopsis thaliana glucosinolate accumulation. Plant Physiol.
126:811–825.
Kliebenstein, D. J., A. Figuthand T. Mitchell-Olds, 2002
Ge-netic architecture of plastic methyl jasmonate responses in
Ara-bidopsis thaliana.Genetics161:1685–1696.
Kliebenstein, D. J., J. Kroymannand T. Mitchell-Olds, 2005 The
glucosinolate-myrosinase system in an ecological and
evolution-ary context. Curr. Opin. Plant Biol.8:264–271.
Kreitman, M., 2000 Methods to detect selection in populations with
applications to the human. Annu. Rev. Genomics Hum. Genet.1:
539–559.
Kuittinen, H., and M. Aguade´, 2000 Nucleotide variation at the
CHALCONE ISOMERASE locus inArabidopsis thaliana.Genetics
155:863–872.
Kumar, S., K. Tamura, I. B. Jakobsenand M. Nei, 2001 MEGA2:
molecular evolutionary genetics analysis software for
microcom-puters. Bioinformatics17:1244–1245.
Lambrix, V., M. Reichelt, T. Mitchell-Olds, D. J. Kliebenstein
and J. Gershenzon, 2001 The Arabidopsis epithiospecifier
protein promotes the hydrolysis of glucosinolates to nitriles
and influences Trichoplusia ni herbivory. Plant Cell13:2793–
2807.
Lazzaro, B. P., 2005 Elevated polymorphism and divergence in the
class C scavenger receptors ofDrosophila melanogasterandD.
Lazzaro, B. P., and A. G. Clark, 2003 Molecular population
genet-ics of inducible antibacterial peptide genes inDrosophila
mela-nogaster.Mol. Biol. Evol.20:914–923.
Little, T. J., J. K. Colbourneand T. J. Crease, 2004 Molecular
evo-lution of Daphnia immunity genes: polymorphism in a gram-negative binding protein gene and an alpha-2-macroglobulin
gene. J. Mol. Evol.59:498–506.
Liu, G., E. B. Holub, J. M. Alonso, J. R. Eckerand P. R. Fobert,
2005 An ArabidopsisNPR1-like gene,NPR4, is required for
dis-ease resistance. Plant J.41:304–318.
Mauch-Mani, B., and A. J. Slusarenko, 1996 Production of salicylic
acid precursors is a major function of phenylalanine
ammonia-lyase in the resistance ofArabidopsistoPeronospora parasitica.Plant
Cell8:203–212.
Maynard-Smith, J., and J. Haigh, 1974 The hitch-hiking effect of a
favorable gene. Genet. Res.23:23–35.
McConn, M., and J. Browse, 1996 The critical requirement for
li-nolenic acid is pollen development, not photosynthesis, in an
Arabidopsismutant. Plant Cell8:403–416.
McDonald, J. H., and M. Kreitman, 1991 Adaptive protein
evolu-tion at theAdhlocus inDrosophila.Nature351:652–654.
McDowell, J. M., A. Cuzick, C. Can, J. Beynon, J. L. Danglet al.,
2000 Downy mildew (Peronospora parasitica) resistance genes
in Arabidopsis vary in functional requirements for NDR1, EDS1, NPR1, and salicylic acid accumulation. Plant J.22:523– 529.
Myers, S. R., and R. C. Griffiths, 2003 Bounds on the minimum
number of recombination events in a sample history. Genetics
163:375–394.
Naur, P., B. L. Petersen, M. D. Mikkelsen, S. Bak, H. Rasmussen
et al., 2003 Oximes in the biosynthesis of glucosinolates in
Arabidopsis. Plant Physiol.133:63–72.
Nawrath, C., S. Heck, N. Parinthawony and J.-P. Me´ traux,
2002 EDS5, an essential component of salicylic acid-dependent
signaling for disease resistance inArabidopsis, is a member of the
MATE transporter family. Plant Cell14:275–286.
Nei, M., 1973 Analysis of gene diversity in subdivided populations.
Proc. Natl. Acad. Sci. USA70:3321–3323.
Nei, M., and W.-H. Li, 1979 Mathematical model for studying
ge-netic variation in terms of restriction endonucleases. Proc. Natl.
Acad. Sci. USA76:5269–5273.
Nei, M., and T. Gojobori, 1986 Simple methods for estimating the
numbers of synonymous and nonsynonymous nucleotide
substi-tutions. Mol. Biol. Evol.3:418–426.
Nordborg, M., and S. Tavare´, 2002 Linkage disequilibrium: what
history has to tell us. Trends Genet.18:83–90.
Nordborg, M., T. T. Hu, Y. Ishino, J. Jhaveri, C. Toomajianet al.,
2005 The pattern of polymorphism inArabidopsis thaliana.PLoS
Biol.3(7): e196.
Olsen, K. M., A. Womack, A. R. Garrett, J. I. Suddithand M.
Purugganan, 2002 Contrasting evolutionary forces in the
Ara-bidopsis thalianafloral development pathway. Genetics160:1641– 1650.
Oppenheimer, D. G., P. L. Herman, S. Sivakumaran, J. Eschand
M. D. Marks, 1991 Amyb-related gene required for leaf
tri-chome differentiation in Arabidopsis is expressed in stipules.
Cell67:483–493.
Pieterse, C. M. J., and L. C. VanLoon, 2004 NPR1: the spider in the
web of induced resistance signaling pathways. Curr. Opin. Plant
Biol.7:456–464.
Purugganan, M. D., and S. R. Wessler, 1994 Molecular evolution
of the plant R regulatory gene family. Genetics138:849–854.
Rausher, M. D., R. E. Millerand P. Tiffin, 1999 Patterns of
evo-lutionary rate variation among genes of the anthocyanin
biosyn-thetic pathway. Mol. Biol. Evol.16:266–274.
Reichelt, M., P. D. Brown, B. Schneider, N. J. Oldham, E. Stauber
et al., 2002 Benzoic acid glucosinolate esters and other
glu-cosinolates from Arabidopsis thaliana. Phytochemistry59: 663–
671.
Ryals, J. A., U. H. Neuenschwander, M. G. Willits, A. Molina,
H.-Y. Steiner et al., 1996 Systemic acquired resistance. Plant
Cell8:1809–1819.
Schlenke, T. A., and D. J. Begun, 2003 Natural selection drives
Dro-sophila immune system evolution. Genetics164:1471–1480.
Schmid, K. J., S. Ramos-Onsins, H. Ringys-Beckstein, B. Weisshaar
and T. Mitchell-Olds, 2005 A multiple sequence survey in
Arabidopsis thalianareveals a genomewide departure from a
neu-tral model of DNA sequence polymorphism. Genetics169:1601–
1615.
Shao, F., C. Goldstein, J. Ade, M. Stoutemeyer, J. E. Dixonet al.,
2003 Cleavage of ArabidopsisPBS1by a bacterial type III
effec-tor. Science301:1230–1233.
Stahl, E. A., G. Dwyer, R. Mauricio, M. Kreitmanand J. Bergelson,
1999 Dynamics of disease resistance polymorphism at the
RPM1locus ofArabidopsis.Nature400:667–671.
Staskawicz, B. J., M. B. Mudgett, J. L. Dangland J. E. Galan,
2001 Common and contrasting themes of plant and animal
dis-eases. Science292:2285–2289.
Sticher, L., B. Mauch-Maniand J. P. Me´ traux, 1997 Systemic
ac-quired resistance. Annu. Rev. Phytopathol.35:235–270.
Stotz, H. U., B. R. Pittendrigh, J. Kroymann, K. Weniger,
J. Fritscheet al., 2000 Induced plant defense responses against
chewing insects. Ethylene signaling reduces resistance of Arabi-dopsis against Egyptian cotton worm but not Diamondback
moth. Plant Physiol.124:1007–1017.
Tajima, F., 1989 Statistical method for testing the neutral mutation
hypothesis by DNA polymorphism. Genetics123:585–595.
Thomma, B. P., I. A. Penninckx, W. F. Broekaertand B. P. Cammue,
2001 The complexity of disease signaling inArabidopsis.Curr.
Opin. Immunol.13:63–68.
Tian, D., H. Araki, E. Stahl, J. Bergelson and M. Kreitman,
2002 Signature of balancing selection inArabidopsis. Proc. Natl.
Acad. Sci. USA99:11525–11530.
Tian, D., M. B. Traw, J. Q. Chen, M. Kreitmanand J. Bergelson,
2003 Fitness costs ofR-gene-mediated resistance inArabidopsis
thaliana. Nature423:74–77.
Toomajian, C., T. T. Hu, M. J. Aranzana, C. Lister, C. Tang,et al.,
2006 A nonparametric test reveals selection for rapid flowering
in the Arabidopsis genome. PLoS Biol.4:e137.
Traw, M. B., J. Kim, S. Enright, D. F. Cipolliniand J. Bergelson,
2003 Negative cross-talk between salicylate- and
jasmonate-mediated pathways in the Wassilewskija ecotype of Arabidopsis
thaliana.Mol. Ecol.12:1125–1135.
Van derBiezen, E., and J. D. G. Jones, 1998 Plant disease-resistance
proteins and the gene-for-gene concept. Trends Biochem. Sci.
23:454–456.
Varet, A., B. Hause, G. Scheeland J. Lee, 2003 TheArabidopsis
NHL3gene encodes a plasma membrane protein and its
overex-pression correlates with increased resistance toPseudomonas
syrin-gaepv. tomato DC3000. Plant Physiol.132:2023–2033.
Veronese, P., M. T. Ruiz, M. A. Coca, A. Hernandez-Lopez, H. Lee
et al., 2003 In defense against pathogens. Both plant sentinels
and foot soldiers need to know the enemy. Plant Physiol.131:
1580–1590.
Voight, B. F., S. Kudaravalli, X. Winand J. K. Pritchard, 2006 A
map of recent positive selection in the human genome. PLoS
Biol.4(3): e72.
Xie, D. X., B. F. Feys, S. James, M. Nieto-Rostroand J. G. Turner,
1998 COI1: an Arabidopsis gene required for
jasmonate-regulated defense and fertility. Science280:1091–1094.
Zhang, Y., W. Fan, M. Kinkema, X. Liand X. Dong, 1999
Interac-tion ofNPR1with basic leucine zipper protein transcription
fac-tors that bind sequences required for salicylic acid induction of
thePR-1gene. Proc. Natl. Acad. Sci. USA96:6523–6528.
Zheng, M. S., H. Takahashi, A. Miyazaki, H. Hamamoto, J. Shah
et al., 2004 Up-regulation ofArabidopsis thaliana NHL10in the hypersensitive response to Cucumber mosaic virus infection and in senescing leaves is controlled by signaling pathways that
differ in salicylate involvement. Planta218:740–750.
Zhou, N., T. L. Tootle, F. Tsui, D. F. Klessigand J. Glazebrook,
1998 PAD4functions upstream from salicylic acid to control
defense responses in Arabidopsis. Plant Cell10:1021–1030.
Zhou, N., T. L. Tootleand J. Glazebrook, 1999 ArabidopsisPAD3,
a gene required for camalexin biosynthesis, encodes a putative
cytochrome P450 monooxygenase. Plant Cell11:2419–2428.
