Pathogen resistance genes represent some of the most abundant and diverse gene families found within plant genomes. However, evolutionary mechanisms generating resistance gene diversity at the genome level are not well understood. We used the complete Arabidopsis thaliana genome sequence to show that most duplication of individual NBS-LRR sequences occurs at close physical proximity to the parent sequence and generates clusters of closely related NBS-LRR sequences. Deploying the statistical strength of phylogeographic approaches and using chromosomal location as a proxy for spatial location, we show that apparent duplication of NBS-LRR genes to ectopic chromosomal locations is largely the consequence of segmental chromosome duplication and rearrangement, rather than the independent duplication of individual sequences. Although accounting for a smaller fraction of NBS-LRR gene duplications, segmental chromosome duplication and rearrangement events have a large impact on the evolution of this multi-gene family. Intergenic exchange is dramatically lower between NBS-LRR sequences located in different chromosome regions as compared to exchange between sequences within the same chromosome region. Consequently, once translocated to new chromosome locations, NBS-LRR gene copies have a greater likelihood of escaping intergenic exchange and adopting new functions than do gene copies located within the same chromosomal region. We propose an evolutionary model that relates processes of genome evolution to mechanisms of evolution for the large, diverse, NBS-LRR gene family.
IN the dynamic interaction between plants and pathogens, plants counter the evolution of pathogen avirulence genes with the evolution of new resistance genes (Flor 1956; Clay and Kover 1996; Stahlet al. 1999). Although our understanding of the molecular pathways and the genomic architecture underlying pathogen resistance traits has greatly advanced over the last decade (Bakeret al. 1997; Leisteret al. 1998; Meyerset al. 1999; Grubeet al. 2000; Panet al. 2000), the genome-level mechanisms generating resistance gene diversity over relatively short time scales are not well understood. To address this problem, we borrowed from phylogeographic approaches (Avise 2000) to determine the evolutionary mechanisms by which numerous and diverse pathogen resistance genes have become dispersed across plant genomes. In this article, we focus on genes encoding proteins containing nucleotide binding site and leucine rich repeat domains (NBS-LRR) because these are the most common class of the resistance genes thus far characterized within plant genomes (reviewed in Bakeret al. 1997; Hulbertet al. 2001). Like many resistance and defense genes, members of the NBS-LRR gene family are found widely distributed throughout plant genomes (Leisteret al. 1998; Meyerset al. 1999; Grubeet al. 2000; Wanget al. 2001) but are clustered within discrete chromosomal regions. Within such clusters, diverse NBS-LRR genes often control recognition of evolutionarily diverse pests and pathogens (Meyerset al. 1999; Gebhardt and Valkonen 2001). To account for the diversity of resistance genes, it has been suggested that resistance gene families evolve by a birth-death process rather than in a concerted fashion (Michelmore and Meyers 1998). Further, several authors have proposed that sequence similarity and function of orthologous resistance genes is conserved because sequence exchange between alleles occurs more frequently than sequence exchange between paralogous gene copies (Michelmore and Meyers 1998; Noëlet al. 1999; Parniske and Jones 1999).
While molecular mechanisms generating variation in NBS-LRR gene copies within chromosomal clusters are becoming better understood, controversy surrounds models of the evolution of diverse NBS-LRR genes and their observed distribution within plant genomes. In this study, we used the complete sequence of Arabidopsis thaliana to test two distinct models for the evolution of NBS-LRR genes on the genomic level. Phylogenetic and comparative mapping studies of resistance gene analogs within the grasses and other plant species suggest a rapid rearrangement model where frequent duplication of individual resistance genes and transposition to ectopic chromosomal locations scatters closely related gene copies across the genome, reminiscent of mobile genetic elements (Gallegoet al. 1998; Leisteret al. 1998; Dubcovskyet al. 2001; Richlyet al. 2002). In contrast, comparative mapping studies within the family Solanaceae provide evidence for a conserved synteny model in which duplication of NBS-LRR genes to nonsyntenic regions is uncommon (Grubeet al. 2000; Panet al. 2000). The conserved synteny and rapid rearrangement models are not mutually exclusive but are used here as a heuristic tool to generate testable predictions for the distribution of paralogous NBS-LRR sequences across plant genomes. The rapid rearrangement model predicts that closely related NBS-LRR sequences, the products of recent gene duplication events, will be found across ectopic chromosomal locations. The conserved synteny model predicts that closely related NBS-LRR genes will be located near each other on the same chromosome and that relocation of NBS-LRR gene copies to new chromosome regions is the result of genome-level evolutionary processes such as polyploid events, chromosome doubling, or recombination between overlapping inversions. The genomes of many plant species such as soybean (Grantet al. 2000), maize (Wilsonet al. 1999), and A. thaliana (Visionet al. 2000) consist of several duplicated chromosomal regions.
We test the fit of data to the differing predictions presented by the conserved synteny and rapid rearrangement model by reconstructing the history of NBS-LRR gene duplication and rearrangement within the A. thaliana genome. Our data set and resulting gene tree are much the same as that of Richly et al. (2002), but in this article we draw on the statistical power of phylogenetic and phylogeographic approaches to examine the correlation between the phylogenetic relationship of NBS-LRR sequences and their physical location in the genome. We find that most ectopic relocation of NBS-LRR gene copies is almost entirely explained by duplication and rearrangement of the chromosome segments on which those genes reside, rather than the independent duplication of individual sequences as inferred in several reports (Leisteret al. 1998; Panet al. 2000; Richlyet al. 2002). Further, we demonstrate for the first time the impact of physical chromosomal location on the frequency of intergenic exchange between NBS-LRR gene copies. We develop a genome-wide model for the evolution of NBS-LRR genes in A. thaliana by bringing together the relationship of NBS-LRR gene duplication, segmental chromosomal duplication, and levels of intergenic recombination events.
MATERIALS AND METHODS
NBS-LRR genealogical reconstruction: We extracted 149 NBS-LRR gene sequences from the MIPS A. thaliana (Columbia) database (http://www.mips.biochem.mpg.de/proj/thal/db/index.html). The NBS-LRR sequences include both those encoding the Toll interleukin receptor motif (TIR genes) and sequences not encoding the TIR motif (non-TIR genes). Except for the region encompassing the TIR, the N-terminal region through the NBS region of NBS-LRR genes is fairly well conserved and suitable for describing evolutionary relationships among diverse sequences. We aligned the amino acid sequence of this region using an iterative hidden Markov model (HMM) algorithm found in the program HMMPRO (Baldiet al. 1994; http://www.netid.com). Gap characters inserted in the region of the TIR motif to achieve alignment were subsequently removed and an alignment of 425 amino acid sites resulted. The genealogy of all 149 NBS-LRR sequences was generated using the Bayesian likelihood program, Mr. Bayes (Huelsenbeck 2000), to create a 50% majority rule consensus tree of 8000 estimated maximum likelihood (ML) trees. Statistical support for the topology of the tree was evaluated using PAUP *4.0b8 (Swofford 2001) to calculate the frequency of branches present among the 8000 ML trees. Branches present in <50% of these trees were collapsed to polytomies. The consensus tree (Figure 1) was midpoint rooted at the ancient division of TIR genes and non-TIR genes (Meyerset al. 1999). The physical position of each NBS-LRR sequence within the A. thaliana genome was determined from the The Arabidopsis Information Resource website (http://www.arabidopsis.org; Hualaet al. 2001) and mapped onto the consensus tree as an unordered character state using MacClade 4.0 (Maddison and Maddison 2000). Phylogenetic clades (clades A–K, Figure 1) were delimited to include NBS-LRR sequences similar enough to allow full-length nucleotide sequence alignment [average pairwise distances within a clade were 0.57–1.57 substitutions per amino acid site; Felsenstein (2001)] but such that each clade also contains sequences located on at least two different chromosomes.
Genome location of NBS-LRR genes and of duplicated chromosome segments: The relative frequency at which NBS-LRR genes duplicate to syntenic or to nonsyntenic genomic regions was estimated by analyzing changes in chromosome location state reconstructed at each node of the genealogy using MacClade 4.0 (Maddison and Maddison 2000). Genome location of NBS-LRR sequences was first defined by their chromosomal location and next, by their location within clusters of NBS-LRR genes located in close physical proximity on the chromosome. In our analyses, chromosomal clusters of NBS-LRR genes were defined by the occurrence of sequences within 2 Mb of one another (2-Mb linked clusters) and are largely the same as those evident in Meyers et al. (1999; http://www.niblrrs.ucdavis.edu/At_RGenes/).
We used the previously identified breakpoints of segmentally duplicated regions within the A. thaliana genome to determine which NBS-LRR gene sequences were found within duplicate regions of the genome (Arabidopsis Genome Initiative 2000; Blancet al. 2000; Grantet al. 2000; Visionet al. 2000). For our purposes, breakpoints of segmentally duplicated regions were defined by the position of the most telomeric and centromeric genes found within the segment. Because the breakpoints of more divergent duplicated regions are difficult to identify by the similarity searches used in the above studies, we accommodated uncertainty in breakpoints by including NBS-LRR genes falling within 0.5 Mb of estimated breakpoints. Further, we used comparative maps between A. thaliana and Brassica nigra (Lagercrantzet al. 1998) to determine if NBS-LRR genes presently found in different regions of the A. thaliana genome might have once been clustered together within an ancestral genome. In these comparative maps, breakpoints were estimated using marker sequences mapped in both A. thaliana and B. nigra genomes, and as above, NBS-LRR genes were included within a chromosome segment if they fell within 0.5 Mb of the mapped sequence in the A. thaliana genome.
Association test for the role of segmental duplication and rearrangement: We employed an index of association test (IA; Brownet al. 1980; Agapow and Burt 1999; http://www.bio.ic.ac.uk/evolve/software/multilocus) to provide a second test of the data's fit to the conserved synteny or rapid rearrangement models. The IA is a summary statistic used to describe the extent of multilocus associations of alleles across populations. For the purpose of the IA test, we treated 2-Mb linked clusters as populations and plotted the occurrence of distantly related NBS-LRR sequences representing different phylogenetic clades (Figure 1, A–K) within these clusters and across chromosomal locations. Sequences representing the different phylogenetic clades were recorded as a single data point within the 2-Mb linked clusters. This approach also limited the effect of variation in numbers of NBS-LRR genes within the same 2-Mb linked cluster on IA test significance values. The statistical significance of observed IA values was obtained by comparing the observed IA values with a null distribution of association values derived from 1000 permutations of NBS-LRR sequence locations across 2-Mb linked clusters. The permutations serve to create expectations for association of gene copies by chance. Sensitivity analyses were conducted by defining smaller clusters of NBS-LRR genes on the basis of sequences located within 0.5 Mb (0.5-Mb linked clusters) of one another and modifying the number of sequences included within a phylogenetic clade.
Intergenic sequence exchange: We examined nucleotide alignments of NBS-LRR genes for evidence of intergenic sequence exchange. The full-length DNA sequences of NBS-LRR genes, both with and without introns, were aligned using CLUSTALW (Thompsonet al. 1994). If lengths of aligned sequences varied greatly such that large gaps were introduced into the alignment, the shortest sequence was removed from the data set and the sequences were realigned. All nucleotide sites in the resulting alignments were examined with GENECONV (Sawyer 2000), which searches for tracts of shared, phylogenetically informative sites between three or more aligned sequences and thus detects sequence exchange events resulting from reciprocal exchange, unequal crossover, or gene conversion events. Only intergenic exchange events maintaining global Bonferroni corrected P values ≤0.05 were considered significant. Sequence mismatch penalties were varied at 0, 1, and 2 to provide the range of detected intergenic exchange events shown in Tables 1 and 2.
Genome location of NBS-LRR gene duplication events: The relative frequency at which NBS-LRR gene duplication occurs to syntenic or to nonsyntenic regions of the genome was estimated by counting the changes of chromosome location state reconstructed at each node of the genealogy and dividing by the total number of nodes involved (Figure 1). Since the A. thaliana sequence represents a single haploid genome, each node within the NBS-LRR genealogy represents a gene duplication event. Thus, we counted each character state change in chromosomal location as duplication to a new location. As a conservative test, we counted only those nodes at which changes of location state or no changes of location state were unambiguously assigned by using MacClade 4.0 (Maddison and Maddison 2000). Implicitly, our approach uses the branches with >50% representation on the consensus tree as an estimate of support for character state changes at each node. An alternative approach using the 8000 ML trees to generate confidence intervals on the inferences of chromosomal location state changes is worthy of development but beyond the scope of our work.
First, we conducted analyses to estimate the fraction of duplication events moving gene copies to the same or different chromosomes. We found that 81.1% of NBS-LRR duplications (90 of 111 reconstructed character states) resulted in gene copies located on the same chromosome as the most closely related paralogous sequence. To further refine the analysis, duplication within and between 2-Mb linked clusters was examined by assigning location within 2-Mb linked clusters as a character state of each sequence and counting changes in character state reconstructed at each node with MacClade 4.0 (Maddison and Maddison 2000). Even using more restricted chromosomal regions, we found that 79.8% of duplication events (71 of 89 reconstructed character states) occur close to one another and within the same 2-Mb linked cluster. These data demonstrate a strong correlation between a close phylogenetic relationship and physical proximity of NBS-LRR gene copies, an observation expected under the model of conserved synteny.
Role of segmental duplication and rearrangement: The majority of observations from our genealogical analyses support the conserved synteny model but do not fully exclude the rapid rearrangement model because some closely related NBS-LRR genes were also found in different regions of the A. thaliana genome (Figure 1). Of the 89 duplication events resolved in the phylogenetic analyses, 18 apparently involved duplication to a new, ectopic chromosomal location. Infrequent duplication of individual sequences to a new chromosome location coupled with more frequent duplication within a region could result in much the same distribution of NBS-LRR sequences as we present above. However, the A. thaliana genome has endured multiple chromosome duplication and rearrangement events, suggesting that segmental duplication and rearrangement might have distributed NBS-LRR gene copies across different chromosome regions (Lagercrantz 1998; Kochet al. 1999; Arabidopsis Genome Initiative 2000; Blancet al. 2000; Grantet al. 2000; Visionet al. 2000).
Using previously identified breakpoints of segmentally duplicated regions within the A. thaliana genome (Lagercrantz 1998; Arabidopsis Genome Initiative 2000; Blancet al. 2000; Grantet al. 2000; Visionet al. 2000), we determined that 15 of the 18 apparent ectopic duplications also fell within segmentally duplicated regions of the genome. Only 3 NBS-LRR gene duplication events of 89 total events are left unexplained by the conserved synteny model and could possibly reflect independent duplication and rearrangement of NBS-LRR genes as predicted by the rapid rearrangement model. Thus, we show that almost all NBS-LRR duplication events involving apparent relocation to a new chromosomal region are explained by segmental duplication and rearrangement. Taking the results of phylogenetic analyses and segmental duplication together, data strongly support the conserved synteny model because most NBS-LRR duplication events (96.6% or 86 of 89 total) occur within local chromosomal regions.
Association test for the role of segmental duplication and rearrangement: Our estimates for the contribution of segmental duplication and rearrangement to the genomic distribution of NBS-LRR sequences are based on the reconstruction of recent duplication events on the gene tree. These inferences will be limited by gene loss, intergenic sequence exchange, and ambiguity in assigning character state changes. As a second test of the conserved synteny and rapid rearrangement models, independent of tree topology, we used an IA test (see materials and methods) to examine the correlated occurrence of NBS-LRR gene copies across the genome. The conserved synteny model predicts that duplication and rearrangement of NBS-LRR gene sequences to new chromosomal locations is accomplished by the duplication and rearrangement of entire chromosomal segments on which those NBS-LRR genes reside. If this is true, we will observe the co-occurrence of gene copies across several chromosome locations. For example, if genes A and B are represented together on one chromosomal segment, their gene copies, A′ and B′, will be represented together on another, duplicate chromosomal segment. In contrast, the rapid rearrangement model predicts that individual NBS-LRR sequences duplicate to new chromosome locations independently of other gene sequences. As a result, genomic locations of distantly related gene copies will not be correlated with each other or with segmental duplications. Note, however, that the ectopic duplication of a single NBS-LRR sequence, followed by rapid duplication within a chromosome location, could generate significant IA values if these recent duplicates were included in the analysis. To avoid confounding recent gene duplication with correlated dispersal, we examined only the association of more distantly related NBS-LRR gene copies. We plotted the genomic location of more distantly related NBS-LRR sequences representing different phylogenetic clades (Figure 1, A–K) onto their sequence position within the A. thaliana genome, with each occurrence within 2-Mb linked clusters represented as a single point (Figure 2). From inspection of Figure 2, it is evident that sequences representing different clades co-occur within more than one distinct genomic region. For example, sequences representative of clades D and E are found together in two separate regions of chromosome 1, in one region of chromosome 4, and in two separate regions of chromosome 5. Similarly, sequences representative of clades F and G are found together in regions of chromosomes 2, 3, 4, and 5. The prevalent associations of distantly related NBS-LRR gene copies across 2-Mb linked clusters led to significant IA tests (P ≤ 0.05) and significance levels were not sensitive to varying the size of linked clusters (2- or 0.5-Mb linked clusters) or varying the delimitation of phylogenetic clades on Figure 1. The correlated occurrence of divergent NBS-LRR sequences across different genomic regions strongly supports the hypothesis that segmental chromosome duplication and rearrangement have distributed members of the NBS-LRR gene family throughout the A. thaliana genome and not the independent duplication and rearrangement of individual genes.
Intergenic sequence exchange: We next used GENECONV (Sawyer 2000) to investigate the impact of physical location and sequence similarity on the frequency of intergenic exchange. The global GENECONV analysis looks for shared phylogenetically informative character states between two sequences as evidence of recombination and results are insensitive to varying substitution models. GENECONV may be considered conservative in its estimate of recombination rates but is useful in detecting intergenic recombination between divergent sequences (Posada and Crandall 2001). As above, closely related NBS-LRR genes were defined as belonging to the same phylogenetic clade (Figure 1, A–K) and physical location was defined by 2-Mb linked clusters. We first examined the frequency of exchange between distantly related sequences and could not detect significant sequence exchange events even when we restricted our attention to sequences lying in close chromosomal proximity and used relaxed sequence mismatch parameters.
Subsequent tests for sequence exchange were conducted by comparing more closely related NBS-LRR sequences. The effect of genome location was determined by comparing the frequencies of intergenic exchange events detected for sequences located within and between different 2-Mb linked clusters. Among sequences located within the same 2-Mb linked cluster, many significant intergenic sequence exchange events (P ≤ 0.05) were found regardless of parameters defined. The inclusion of intron sequences provided a larger number of polymorphic sites and increased the number and length of sequence exchange events detected (Table 1). Between closely related NBS-LRR sequences located in different 2-Mb linked clusters, a number of significant sequence exchange events (P ≤ 0.05) were also detected and again, inclusion of intron sequences allowed more events to be detected (Table 2). Considerable variation in the frequency and length of recombination events is evident along chromosomes (Table 1) and across different clades (Table 2). Comparing the two sets of results, the average frequency of exchange detected among sequences residing within the same 2-Mb linked cluster was ∼20-fold higher and involved longer exchange tracts (Table 1) than exchanges between sequences residing in different 2-Mb linked clusters (Table 2). Most intergenic exchange between NBS-LRR sequences on different chromosomes involved sequences located within duplicated genomic regions, suggesting that some of these exchanges might have occurred before segmental duplication or rearrangement events. Mutations and gene loss following a duplication event would obscure the history of sequence exchange and lower the probability of detecting ancestral exchanges. Altogether, our results show that frequency of exchange detected between NBS-LRR sequences from different chromosomal regions is detectable but much lower than the frequency of intergenic exchange observed within chromosomal regions.
Using a novel phylogeographic approach treating chromosomal regions as geographic populations, we provide strong support for a conserved synteny model of NBS-LRR gene evolution in which most NBS-LRR gene duplications occur within restricted chromosomal regions.
Previously, the observation that closely related resistance gene sequences are located across widespread chromosomal locations has been interpreted as support for a model of rapid rearrangement of duplicated resistance genes (Gallegoet al. 1998; Leisteret al. 1998; Panet al. 2000; Dubcovskyet al. 2001; Richlyet al. 2002), as might our evidence for sequence exchange between genes in ectopic locations. In contrast, we derive support for the conserved synteny model by showing that duplication and rearrangement of chromosomal segments, rather than duplication of individual gene sequences, drives the genomic distribution and apparent ectopic duplication of NBS-LRR gene sequences.
Our finding that most recent gene duplication events have occurred within a local chromosomal region is supported by results from comparative mapping and plant genome projects (see Hulbertet al. 2001), which show that several closely related gene copies are clustered together within local chromosomal regions. Such resistance gene clusters may themselves be located within larger chromosomal regions that carry diverse genes encoding resistance to several different pathogens (Gebhardt and Valkonen 2001). Here, we build on previous models for the evolutionary dynamics within NBS-LRR gene clusters (Hulbert 1997; Parniskeet al. 1997; Michelmore and Meyers 1998; Noëlet al. 1999; Parniske and Jones 1999; Chinet al. 2001; Doddset al. 2001; Sunet al. 2001) to develop a genome-level model for the evolutionary dynamics of NBS-LRR genes.
We took advantage of detailed information for segmental duplication and rearrangement in the evolution of the A. thaliana genome (Lagercrantz 1998; Arabidopsis Genome Initiative 2000; Blancet al. 2000; Grantet al. 2000; Visionet al. 2000) to correlate segmental duplication and rearrangement events with gene duplication events. Tracking changes in chromosomal location associated with gene duplication events on the genealogy, we demonstrate that ∼80% of duplication events occur within relatively restricted chromosomal regions while the remaining 20% involve duplication to new chromosomal locations.
Our estimate for the level of local duplication (80%) is larger than that of Richly et al. (2002) who estimated that 50% of gene copies remain in local chromosomal regions. Differences in these estimates arise from two sources. Richly et al. (2002) defined clusters of NBS-LRR sequences as genes occurring within eight open reading frames of each other, regions that may be smaller than the physically delimited distances (2- or 0.5-Mb linked clusters) used in this study. More importantly, the pairwise approach of Richly et al. (2002) considers changes in chromosome location states only at terminal branches of the genealogy. Using character state reconstruction accounts for the inherent correlation of events within a genealogy and thus avoids overestimation of character state changes. Both the results presented here and those of Richly et al. (2002) estimate that the majority of NBS-LRR duplication events occur locally but the studies differ in their attribution of evolutionary mechanisms for genome-wide dispersal of gene copies. The most striking results were obtained when we included additional information concerning the evolutionary history of segmental duplication and rearrangement within the A. thaliana genome (Lagercrantz 1998; Grantet al. 2000; Visionet al. 2000). Without incorporation of genome-level processes, each NBS-LRR gene copy that lands in an ectopic location will be counted as an independent and rapid rearrangement event (Leisteret al. 1998; Richlyet al. 2002). With genome-level data, we find that most all NBS-LRR gene duplications involving ectopic relocation (16% of the total number of duplications) are explained by duplication of the chromosomal segments on which they reside. The conserved synteny model accounts for the genome-wide distribution of 96% of the total NBS-LRR gene duplications.
The conclusions resulting from phylogenetic reconstruction were further supported by results for the co-occurrence of distantly related genes within more than one distinct genomic region. We reasoned that if segmental duplication and rearrangement drive the genomic distribution of NBS-LRR genes, then all genes located on these segments will be duplicated together. As a result, gene copies will occur together in different genome locations more often than expected under independent duplication and rearrangement of individual genes. We treated NBS-LRR gene clusters as populations across the genome and examined the correlated occurrence of NBS-LRR gene copies in these populations using an index of association test. The index of association provides a robust statistical test of whether observed association of sequences in different chromosomal regions deviates significantly from random associations expected under the rapid rearrangement model. The results demonstrate significant associations of genes representing different phylogenetic clades across several genomic locations. The impact of intergenic exchange on the IA results is negligible because measurable levels of intergenic exchange do not occur between divergent sequences representative of different phylogenetic clades (see below). Together, the results of phylogenetic and association tests demonstrate that segmental duplication and rearrangement are the most important mechanisms in distributing NBS-LRR gene family members across the genome.
Intergenic exchange between NBS-LRR genes: Finding that the genomic distribution of NBS-LRR sequences is dependent on segmental duplication and rearrangement has profound implications for mechanisms of molecular evolution in this gene family. In this study, we examined the relationship of sequence similarity and genomic position to the frequency of intergenic exchange between NBS-LRR gene copies. Evidence of numerous sequence exchange events was detected between closely related NBS-LRR genes, confirming the results of Chin et al. (2001) and Sun et al. (2001), while no intergenic exchange events were detected between NBS-LRR sequences representative of different phylogenetic clades, even when those sequences were in proximity to one another. Similarly, the recent study of positive selection on NBS-LRR genes in A. thaliana (Mondragón-Palominoet al. 2002) detected little to no recombination between more distantly related NBS-LRR genes located in different chromosomal regions. Interestingly, the position of recombination events we detected within NBS-LRR coding sequences did not correlate with the codon sites shown by Mondragón-Palomino et al. (2002) to be under positive selection, indicating that convergent evolution did not significantly inflate our estimates of recombination events between NBS-LRR genes.
In our study, the number and length of detected intergenic exchange events varied among chromosomal clusters and phylogenetic clades. Resolution of major clades within the genealogy was not impeded by intergenic exchange but exchange rates are high enough to contribute to homoplasy and phylogenetic grouping of sequences within clades (Wiuf and Hein 1999, 2000; Schierup and Hein 2000), similar to the results of Dodds et al. (2001) for the N locus of flax. However, as previously shown for the major histocompatibility complex (MHC; Takahata and Satta 1998; Martinsohnet al. 1999), we infer that intergenic exchange rates have not been high enough to cause the concerted evolution of NBS-LRR sequences since phylogenetic relationships are resolved and polytomies are not common within clades.
While several studies have examined intergenic exchange between NBS-LRR gene copies within the same chromosome region (Hulbert 1997; Michelmore and Meyers 1998; Chinet al. 2001; Sunet al. 2001), few studies have compared rates of sequence exchange within and between different chromosome regions (see Parniske and Jones 1999). Examining evolution within gene clusters, birth-death models used to describe NBS-LRR (Michelmore and Meyers 1998), and MHC gene evolution (Takahata and Satta 1998; Martinsohnet al. 1999) imply that intergenic exchange is limited and that mutation, interallelic sequence exchange, and divergent selection drive the diversification of gene copies within physical clusters. While the complete genome sequence of Arabidopsis does not allow evaluation of interallelic exchange, we did detect significant intergenic exchange between paralogous NBS-LRR genes within local chromosomal regions and our conclusions contrast somewhat with the Michelmore and Meyers (1998) model. Finding intergenic exchange is not itself surprising, since the rates of intergenic sequence exchange between duplicate genes found in close physical proximity have been estimated well above the neutral mutation rate in plants (Hulbert 1997; Parniskeet al. 1997; Jeleskoet al. 1999; Chinet al. 2001; Doddset al. 2001; Sunet al. 2001). Sequence exchange rates and their effects on NBS-LRR gene function vary widely across different plant systems (Hulbert 1997; Parniskeet al. 1997; Chinet al. 2001; Doddset al. 2001; Sunet al. 2001) and different chromosomal regions (this study) but could contribute to gene structural and functional conservation among members of a physical cluster.
Given our observation of frequent intergenic exchange, the divergence of duplicate NBS-LRR genes within clusters must be rapid to escape relentless sequence homogenization by intergenic exchange (Walsh 1985, 1987). Divergence between paralogous gene copies has been proposed to occur by relaxation of selection on gene copies following duplication (Lynch and Conery 2000) and interallelic recombination at rates greater than intergenic recombination (Michelmore and Meyers 1998). Regardless of mechanisms generating sequence diversity, strong diversifying selection must push the evolution of paralogous gene copies at a rate greater than the rate of conversion or accumulation of mutations to pseudogenes for sequences to escape homogenization by intergenic exchange (Walsh 1985, 1995). Our results do not directly address selection levels but we note that divergent NBS-LRR gene sequences are found in close physical proximity and demonstrate that these experience few exchange events. For example, sequences from clades D and E are always found within close physical proximity. An extreme example is found on the bottom arm of chromosome 5, where sequences representing all clades except J are found together in a “supercluster” (Richlyet al. 2002). These dissimilar and yet clustered genes are likely records of strong diversifying selection (Meyerset al. 1999; Mondragón-Palominoet al. 2002).
Effect of segmental duplication on the rates of intergenic exchange: Segmental duplication and rearrangement account for a small proportion of NBS-LRR gene duplications but put gene copies into a new evolutionary context, out of the reach of intergenic exchange. We present the first empirical evidence validating models of Ohta and Dover (1983), which predicted lower levels of intergenic exchange rates between gene copies located on different chromosomes, as we show that intergenic sequence exchange is much more limited among duplicate NBS-LRR sequences located in different chromosomal regions than among sequences located within the same region. The limited intergenic exchange that we inferred between NBS-LRR genes in different genomic regions could result from intergenic exchange that occurred before segmental duplication or actual intergenic exchange between NBS-LRR genes in different genomic locations. Nevertheless, as Ohta and Dover realized, distributing duplicate sequences to different chromosomal regions reduces the introduction of new sequence into a cluster of gene copies by intergenic exchange. With lower levels of sequence exchange between gene clusters, divergence between sequences in different clusters will accelerate relative to variation within clusters. Given the frequencies of exchange that we estimate within and between chromosomal clusters, Ohta and Dover's (1983) model predicts that the probability of identity between gene copies located in different genomic regions will be at least fourfold lower than identity among sequences within clusters. Segmental duplication and rearrangement will allow duplicate sequences to escape homogenization of intergenic exchange under selection levels lower than would be necessary for escape from homogenization within a cluster (Walsh 1987).
A model for the genomic evolution of the NBS-LRR gene family: We present a genome-level model in which the evolutionary trajectory of NBS-LRR genes is primarily governed by two mechanisms. First, most of the dynamic variation in NBS-LRR gene copy number occurs within local chromosomal regions. New NBS-LRR genes can arise and be lost through unequal crossing over, conversion, and an accumulation of mutations leading to either a pseudogene or a new function (Walsh 1995; Michelmore and Meyers 1998; Lynch and Force 2000). Infrequently, new resistance specificities evolve and are maintained through time by balancing selection (Songet al. 1997; Stahlet al. 1999; Stahl and Bishop 2000) against the grind of homogenization and gene loss. Second, although accounting for a smaller fraction of gene duplication events, segmental duplication will have an impact on NBS-LRR gene family diversification not previously recognized. NBS-LRR gene copies in new chromosomal locations will more often escape intergenic exchange and adopt new functions under levels of selection lower than required for gene copies within a cluster. Further, segmental duplication could allow the preservation of many alleles that would otherwise be maintained at a single NBS-LRR locus (Otto and Young 2002). The distribution of NBS-LRR sequences across the A. thaliana genome is the result of a dynamic history of gene duplication within chromosomal regions coupled with segmental duplication and rearrangement of larger chromosomal regions. The genome-level model presented here describes the relationship between genomic architecture and evolutionary processes in an important gene family and thus provides a model to be tested in other systems and compared to the findings for A. thaliana.
We thank Anja Forche, James Garton, Andrew Munkasci, and Ronald Phillips for many helpful discussions and Christine Ramos for computer assistance. Research was supported by a National Science Foundation Plant Genome Initiative grant to G. May (DBI-9975866), a Plant Molecular Genetics Institute graduate fellowship to A. Baumgarten, and a U. S. Department of Agriculture National Needs graduate fellowship to S. Cannon.
Communicating editor: J. B. Walsh
- Received August 9, 2002.
- Accepted March 14, 2003.
- Copyright © 2003 by the Genetics Society of America