Grass Genomic Synteny Illuminates Plant Genome Function and Evolution

The crop circle

The first comparative genetic map in the grasses was a miniscule maize::sorghum comparison in a study meant to test whether restriction fragment length polymorphism (RFLP) markers generated in one species (e.g., maize) could be used to help generate a genetic map in other species (e.g., sorghum) [38]. This project indicated that maize RFLP probes could be used routinely for species as far distant as foxtail millet, a lineage that last shared a common ancestor with the maize lineage about 30 million years ago (mya) [48]. Serious grass genome comparisons were then generated by expert mapping labs, especially the Gale and Tanksley groups [3, 4, 20]. These studies indicated a good deal of similarity in gene content and colinearity, with a low frequency of small and large exceptions. Moore et al. [67] provided a major conceptual leap when they identified a series of conserved grass genome segments and then assembled them into a comparative circle map.

The comparative circular map of the grasses, also known as the crop circle, has allowed identification of the major rearrangements that differentiate grass genomes, and has provided insight into the timing of these events during grass descent from a common ancestor more than 50 mya. As shown in Fig. 1, gene order and telomere location are largely conserved at this scale, although the number of chromosomes is quite variable across species. Maize yields two concentric circles, suggestive of a whole genome polyploidy event, which has been confirmed by extensive orthologous DNA sequence analysis [80]. Two specific translocations, shown at 3 o’clock and 7 o’clock on the circle map, are shared by all investigated members of the Panicoid subfamily but not by rice or the Triticeae. Most of the other detected rearrangements are inversions that are limited to only one or two of the species depicted (Fig. 1).

In addition, this circular map allowed the easily visualized (and thus conceptualized and transmitted) discovery that some important genes involved in domestication or other important traits appeared to be the same orthologous loci across multiple grass species (Fig. 1) [72]. This, in turn, helped encourage the use of surrogate plant chromosomes like the relatively small genome of rice to assist in the map-based cloning of genes in large-genome species like barley, wheat, or sugarcane [14, 33, 19].

Three major conclusions that were clear from the comparative circular maps were (1) the relatively low frequency of large genomic rearrangements, (2) the presence of inversions, translocations and duplications, and (3) the uneven distribution of such events, with many at boundaries near current centromere locations (Fig. 1). Among the many issues not resolved by this analysis, however, was whether the frequencies of these major rearrangements were in any way predictive or mechanistically similar to the frequencies and types of local rearrangements. More detailed physical and genetic maps would be needed to address these questions.

Physical maps, genetic maps and their comparison

Early studies by comparative genetic mapping revealed the extent of conservation of gene content and gross gene order among different grass species, but did not give many insights into the likelihood or nature of small rearrangements. In these first studies [38], it was observed that most maize RFLP probes hybridized strongly to sorghum DNA, but the repetitive DNA sequences in maize usually did not hybridize to sorghum. This suggested that repetitive DNA sequences evolved much faster than genes, and that heterologous probes could thus provide some advantages over homologous probes from a repeat-rich genome.

Comparative genetic mapping between closely related grasses, such as sorghum and sugarcane, whose separate lineages diverged from each other about 8 mya, show striking map colinearity [35]. In contrast, detailed comparative genetic mapping among more distantly related species, such as maize and rice, identified numerous chromosomal rearrangements, such as telomeric fusions, nested insertions, inversions and translocations [92], although about 2/3 of these genomes appeared to still be colinear. Many of the detected rearrangements were confirmed by comparative physical mapping, such as (from a rice perspective) the fusion of rice chromosomes 3 and 10 and chromosomes 7 and 9 into single chromosomes in the Panicoideae lineage [88]. In addition, comparative physical mapping also uncovered the ancient grass genome duplication shared by maize [88], wheat [77], and other grasses [71].

Comparative physical mapping between sorghum and rice revealed different genome components with very different degrees of microcolinearity. In euchromatic regions, where most meiotic recombination occurs, greater microcolinearity was observed; however, less microcolinearity was observed in recombination-poor heterochromatin, such as pericentromeric regions [12]. This phenomenon was also apparent in comparison of homoeologous chromosomal regions in rice derived from the ancient duplication at the origin of the grasses, where little colinearity was retained in pericentromeric regions [83]. In addition, the heterochromatic regions of sorghum have been preferentially expanded relative to rice, as compared to euchromatic regions [51]. Future detailed studies of microcolinearity in heterochromatin are needed to uncover the dynamics and mechanisms for macro- and micro-rearrangements in these crossover-deficient parts of grass genomes [61, 63].

Across the grasses (and a bit beyond)

Even from the start, comparisons of genomic sequence in orthologous regions of different grass species examined a very large time frame, such as rice versus sorghum [17] or rice versus various Triticeae [28, 37, 26], all comparisons where the investigated species last shared a common ancestor ~50 mya. In this time frame, the sequences between genes appeared to be completely different, although very tiny “conserved non-coding sequences” (CNS) were later discovered [45, 36]. Even introns of orthologous genes, although largely consistent in location across all flowering plants, contained obvious conserved sequences only at the boundaries needed to specify appropriate RNA processing. Hence, the general conclusion could be reached that anything still conserved after 50 million years of grass genome divergence was likely to have an important function.

Gene content and order, on the other hand, were mostly conserved on segments of a few dozen to a few hundred kb even after 50 million years of independent grass genome evolution. Comparisons to rice have been particularly useful in this regard because (1) it is evolutionarily quite distant from the other important grasses like maize, wheat, barley, and sorghum [48], (2) it has a relatively small genome (~400 Mb) with a high gene density, (3) it’s genome was an early target for comprehensive sequence analysis [41], and (4) it has proven to be more stable vis-à-vis small local rearrangements than other grasses like maize, sorghum, wheat or barley [9].

In the most comprehensive comparisons to date, between rice and two panicoid grasses, sorghum and maize, the frequency of gene movement over the last fifty million years was calculated as at least 5%, and possibly as high as 25%, between sorghum and rice [53]. This number does not include the gain or loss of tandemly repeated gene copies, a very common phenomenon in all grass lineages investigated. Most of the genic rearrangements in maize compared to either rice or sorghum are apparent gene losses on one of two maize homoeologues [40, 53], an expected outcome of the polyploidization event about five mya that gave rise to the Zea lineage [80]. However, too little data yet exist to identify possible subtle patterns in types of rearrangement. Moreover, rearrangements involving genes are likely to be under selective pressure, so the events currently observed in any species are a combined outcome of those events that have occurred, minus those that were subsequently removed by chance or by selection against some specific changes.

In more distant comparisons, with longer ancestral divergence times, colinearity across orthologous regions appears to be much more rare than within the grasses. In the rice flatsedge, Cyperus iria, the near-adjacent Sh2 and A1 homologues appear to be conserved in order and orientation, but one of the two genes in between in the grasses is missing in the sedge (A. Pontaroli and J. Bennetzen, unpub. obs.). However, this is the only comparison that has been done to the grasses in this ~110-million-years-of-divergence window [13]. Similarly, Musa (e.g., banana) genomes show some colinearity with the grasses after >115 million years of divergence from their last shared ancestor, but more than 50% of the annotated genes were non-colinear in a comparison to rice [56]. With even more distant comparison to the eudicots, >220 million years of independent descent, only rare segments of genic colinearity are observed at either full genome or local genome scales [58].

The most frequent type of structural change in all investigated angiosperm nuclear genomes has been observed to be the differential insertion and subsequent instability of transposable elements (TEs). In large-genome species like maize and barley, most of the DNA between genes is comprised of TEs, especially long terminal repeat (LTR) retrotransposons [78, 86, 89, 75]. These elements transpose by reverse transcription of an RNA transcript and insertion of the resultant DNA, so transposition does not involve excision. Because LTR retrotransposons make up more than 50% of most or all large flowering plant genomes and their high content varies somewhat proportionally with angiosperm genome size, it is clear that these TEs are the most important factor responsible for genome size variation in flowering plants [10]. Because these TEs (and all other unselected DNAs) are fragmented and removed so rapidly by accumulated small deletions (see below), all of the insertions appear to be very recent, usually within the last 2–6 million years [87]. This accounts for the near-complete lack of homology of the intergenic regions in orthologous genome segments with grass lineages that last shared a common ancestor more than 50 mya.

We currently lack a vocabulary to precisely describe the degree of conservation of genic content and colinearity between any two species, much less across multiple species, although a gene-pair conservation terminology is currently in development (L. Feng and J. Bennetzen, unpub. res.). However, it is clear that some lineages are very unstable (e.g., pearl millet, sorghum, maize) and others are much more stable (e.g., rice and foxtail millet) at the level of compared genetic maps and/or microcolinearity [23, 75, 9, 40]. We do not yet know the reasons for these differences, nor whether high conservation at one scale (e.g., genetic map) in any way correlates with high conservation at other scales (e.g., physical map or microcolinearity). It is clear, though, that certain types of gene rearrangement are rare (e.g., movement of a gene to a wholly different chromosome) while others are relatively common (tandem duplication, deletion or inversion of small genic segments).

Analysis of microcolinearity and gene content conservation at long time frames has the advantage of the accumulation of multiple events for analysis, but this is more than counterbalanced by three negative aspects of concentrating on such ancient rearrangements. First, natural selection has had a great deal of time to remove any events that had even a minor organismal disadvantage, so one only observes certain classes of tolerated or advantageous events that might not be proportional to the true spectrum of de novo rearrangements. Second, the components of the genome responsible for the rearrangement have had ample time to decay into a state where they are invisible to current annotation approaches. And, third, individual events may be buried underneath second, third or more layers of events at the same location. For all of these reasons, investigations of orthologous regions in closely related lineages are justified, and are expected to be “there to discover” because of the relatively high rate of local chromosomal rearrangement in the grasses.

Colinearity dynamics within a 0–15 million year window of grass genome evolution

Orthologous sequence comparisons across short time frames has the potential to reveal both the rate and the mechanisms for disruption of colinearity. In a sequence comparison of the adh1-orthologous regions of maize and sorghum, two species that last shared a common ancestor about 12 mya [80], a 212-kb maize sequence was found to be largely collinear with a 66-kb sorghum sequence [84]. The more than three-fold size difference is mainly due to nested LTR retrotransposon insertions in the maize genome [78, 84]. In the original annotation, orthologs of nine maize genes were detected in the sorghum region in perfect colinear order; however, three additional genes in this sorghum segment were not found in the maize adh1 region. In subsequent analyses, one of the “missing” maize genes was found to be located in the adh1-homoeologous region of maize [40]. This has now turned out to be a routine situation in the maize genome, where two maize segments represent each sorghum region due to a polyploidy event in the Zea lineage within the last few million years [80]. Gene deletion (usually of only one homoeologous copy) subsequent to polyploidization has now reduced the originally doubled copy number of genes (2×) to less than 1.5× [53]. The other two non-colinear genes in the adh1-orthologous regions of sorghum are found elsewhere in the genomes of maize and other grasses and are hypothesized to have been caused by the insertion of two unlinked genes, either as two subsequent events or by a single event involving three chromatids. In dramatic contrast, a comparison of the adh1-orthologous regions between sorghum and sugarcane, both gene colinearity and strong homology of non-coding regions were observed [42], indicating greater stability in these lineages over this shorter (~8 million year) time frame of divergence.

In at least some genomes, polyploidization is followed by extensive genomic change resulting in the silencing and elimination of duplicated genes [1]. In grasses, polyploidy has been a recurrent theme, with many lineages exhibiting full genome duplications over the last few million years. Local sequence comparisons in these species, such as maize [40, 53, 64], wheat [27, 46, 90, 34, 16, 15] and sugarcane [42], have revealed interesting features of gene and genome evolution in recent polyploids. LTR retrotransposon amplification and altered regulation (e.g., silencing) or loss of duplicated genes are repeated themes. Inactivation and eventual elimination of duplicated genes can be mediated by altered epigenetic regulation, deletions, TE insertions, and/or point mutations causing premature stop codons.

Some evidence suggests that specific alterations recur in independent polyploidizations in wheat [27, 46] and Brassica napus [59]. However, most eventually fixed changes do not occur instantly in post-polyploid genome rearrangements, at least not in maize. In adh1-homoeologous regions, for instance, fragments of partially deleted genes remain, indicating the incomplete status of removal several million years after polyploidy, and showing that these gene losses are primarily by the accumulation of multiple small deletions [40]. Another example of reasonably stable polyploid gene copies comes from a comparative study of the adh1-orthologous regions of maize, sorghum and sugarcane [42]. The two sugarcane homoeologous haplotypes show perfect genic colinearity. In addition, two maize homoeologous regions yielded the same gene content, order and orientation as in sugarcane. Our data on comparative analysis in the Oryza genus also reveals excellent stability of polyploid genomes formed less than two million years ago (Chen et al., unpub. res.).

The Oryza genus contains about 24 species that belongs to ten different genome types [31]. A project, entitled the Oryza MAP Alignment Project (OMAP), was launched to build a framework for comparative biology in the Oryza genus [93]. Representative species, ranging from closely related species/subspecies, such as those with AA genomes, which diverged from their common ancestor less than a million years ago, to more distantly related species, such as O. brachyantha and O. granulata, whose ancestors diverged about 10 mya, were chosen for bacterial artificial chromosome (BAC) library construction, BAC end sequencing, and physical map construction [5, 49]. The initial analyses revealed excellent gene colinearity both in their physical maps [50] and in sequence comparisons [96]. Genome size variations in the Oryza genus were found to be mainly caused by lineage-specific amplifications of LTR retrotransposons [74, 6]. Our systematic comparative analysis of the sequence of the MONOCLUM1-ortholgous regions across the Oryza genus not only revealed high gene colinearity but also identified new genes that appear to have originated de novo in the AA genomes (Fig. 2 and Chen et al., unpub. res.), which highlights the advantage of multiple species comparisons.

Intraspecific local sequence comparisons have also identified interesting features of grass genome structure and evolution. A detailed sequence comparison of the bronze region of maize inbred lines McC and B73 found that LTR retrotransposon clusters differed one hundred percent in location relative to the genes in the bronze region between these two lines [29]. This suggests an amazingly rapid process both for TE insertion and for removal of ancestral TEs. In addition, the first annotation of these two regions suggested that the genes themselves differed between these lines in this region. An apparent four-gene cluster was detected in McC but not in the orthologous position in B73 [29]. Later, these sequences were found to be comprised of four gene fragments within Helitrons, a new type of eukaryotic transposon [44, 54, 52, 68]. This phenomenon resembles Pack-MULEs, a type of TE first named and comprehensively described in rice, that also capture and mobilize gene fragments [43]. Although neither Helitrons nor Pack-MULEs usually mobilize intact genes, they do commonly acquire more than one gene fragment in the same element. When transcribed, these internal fragments are often fused (via intron processing) into transcripts that could encode novel protein products [43, 68]. This process of exon shuffling, first proposed by Gilbert [32] as the reason for the existence of introns, could be creating new genes in plants at an amazing rate. The maize nuclear genome, for instance, has more than 4,000 Helitrons that contain inserted gene fragments [68] (L. Yang and J. Bennetzen, unpub. res.). However, there is not yet a proven case of any of these Helitron- or Pack-MULE-generated “new” genes having actually acquired a genetic function essential to its host. Given the rapid rate of unselected DNA loss from plant genomes (see below), it is unlikely that conversion of these chimeric gene candidates into true genes will occur commonly, but even rates as low as one in a million would be significant. Other than the standard route of gene duplication, which primarily creates subfunctionalized or (rarely) mildly modified new gene functions (reviewed in [39]), there is no known aggressive process for the generation of new genes. Perhaps Pack-MULEs and Helitrons will eventually be proven to provide this process. At the very least, we expect to see the discovery of more cases of TE components being co-opted for organismal functions in plants, as in the recent identification of transcription factors in Arabidopsis derived from the Mutator transposase [57].

Even if TE-vectored gene fragments are rarely if ever true genes with a selected host function, they certainly are a complication to genome annotation. Even without internal gene fragments, low-copy-number TEs are often mis-annotated as genes, giving rise to as much as two-fold over estimations of gene numbers [7]. This type of over-estimation in gene number can play particular havoc with assessment of genic colinearity, as evidenced by studies in rice showing hundreds of gene differences between different races of O. sativa that were later shown to all be explained by mis-annotated TEs [9]. Hence, many early publications showing numerous genic exceptions to microcolinearity are incorrect because of this routine annotation error.

Sequence comparisons in closely related haplotypes in Arabidopsis, in rice and in wheat have demonstrated that unequal homologous recombination and illegitimate recombination are the major forces that remove DNA from flowering plant genomes [16, 21, 60, 90, 91]. These activities can remove >100 Mb of DNA from a plant genome in just one million years [62], but the rate of removal appears to be much faster in some angiosperms than in others [87]. Most of the removed DNA is derived from TEs, but other intergenic DNA and extra gene copies are also removed by these processes [60].

Several recent studies have accentuated the fact that not all genomic regions evolve at the same rate. Disease resistance gene clusters are known to be unstable even in map position [55], and to also undergo high rates of unequal recombination [76], including some recombination events that are delimited to specific sites that can optimize novel pathogen recognition specificities [70]. Ribosomal RNA gene clusters also appear to vary in map position even in close relatives [25]. Perhaps most surprising, the composition and arrangement of sequences in centromeres have been found to be hyper-variable, primarily by the process of unequal homologous recombination [61, 63, 65]. This rapid rearrangement by recombination in a region that is deficient in crossovers suggests a very tight control over the outcomes of recombination, especially a powerful bias toward non-crossover, intrastrand and/or sister chromatid outcomes [61]. This core centromeric instability has been argued to yield centromeres that have the potential to out-compete other centromeres for choice as the germinal nucleus in egg development [66].

In summary, local sequence comparisons of closely related grass genomes and of intraspecific haplotypes have begun to reveal the major mechanisms driving genome evolution. These include gene and genome duplication, gene silencing and eventual deletion of duplicated genes subsequent to polyploidization, transposable element amplification, gene movement mediated by transposition of mobile elements, unequal homologous recombination, and illegitimate recombination. All of these processes are quite variable even when comparing closely related species, so their differences in levels of activity (and, possibly, specificity) are responsible for the very different genomes found in flowering plants.