|Home | About | Journals | Submit | Contact Us | Français|
We advocate here a 1001 Genomes project for Arabidopsis thaliana, the workhorse of plant genetics, which will provide an enormous boost for plant research with a modest financial investment.
Thale cress, Arabidopsis thaliana, is a member of one of the largest families of flowering plants, the Brassicaceae, to which mustards, radishes and cabbages also belong. A. thaliana is thought to have originated in Central Asia and spread from there throughout Eurasia. During the last glaciation, A. thaliana was confined to the southern limit of its range, and after the ice retreated, much of Europe was recolonized by different populations, resulting in complex admixture patterns. Today, A. thaliana occurs throughout the Northern Hemisphere, mostly in temperate regions, from the mountains of North Africa to the Arctic Circle (Figure (Figure1).1). Like many other European plants, it has also invaded North America, most probably during historic times [1-5].
The ascendancy of A. thaliana to become one of the most popular species in basic plant research , despite its lack of economic value, is due to the favorable genetics of this plant. It has a diploid genome of only about 125 to 150 Mb distributed over five chromosomes, with fewer than 30,000 protein-coding genes. The ease with which it can be stably transformed is unsurpassed by any other multicellular organism . Moreover, as flowering plants only appeared about 100 million years ago, they are all relatively closely related. Indeed, key aspects of plant physiology such as flowering are highly conserved between economically important grasses such as rice and A. thaliana .
A. thaliana was the first plant species for which a genome sequence became available. This initial sequence was from a single inbred strain (accession), and was of very high quality, with each chromosome represented by merely two contigs, one for each arm . In addition to functional analyses, the 120 Mb reference sequence of the Columbia (Col-0) accession proved to be a boon for evolutionary and ecological genetics. A particular advantage in this respect is that the species is mostly self-fertilizing, and most strains collected from the wild are homozygous throughout the genome. This distinguishes A. thaliana from other model organisms such as the mouse or the fruit fly. In these systems, inbred strains have been derived, but they do not represent any individuals actually found in nature.
Natural A. thaliana accessions show tremendous genetic and phenotypic diversity [10,11] (Figure (Figure1b).1b). Over the past 10 years, traditional quantitative trait locus (QTL) mapping has led to the identification of sequence variants that modulate a range of physiological and developmental traits, from germination and flowering to ion content [10,11]. Prior knowledge of the biological function of the affected genes was often helpful in identifying them, but increasingly, the responsible locus is found to encode a protein without known biochemical function such as the FRIGIDA (FRI) flowering regulator or the DELAYED GERMINATION1 (DOG1) gene [12-14]. Apart from alleles that alter expression levels or protein function, a surprising number of drastic mutations such as deletions and stop codons underlie phenotypic variation. Some of these changes are found in many accessions (see, for example [12,15]), suggesting that they are adaptive. Nevertheless, despite some success stories, the number of known alleles responsible for phenotypic variation among accessions remains limited, mostly because fine mapping and dissection of QTLs are so tedious.
Efforts to accelerate the discovery of functionally important variants began with a large-scale study in which some 1,000 fragments across the genomes of 96 accessions gathered from all over the word were compared by dideoxy sequencing . A major conclusion from this work was that there has been considerable global gene flow, so that most sequence variants are found worldwide, although genotypes are not entirely random. There is isolation by distance, and even though population structure is relatively moderate, it can easily be a confounding factor in association studies. These properties are reminiscent of what has been described for humans [16-20].
From this first set of 96 strains, 20 maximally diverse strains were chosen for much denser polymorphism discovery using array-based resequencing . This led to the identification of about one single nucleotide polymorphism (SNP) for every 200 bp of the genome, constituting one quarter or so of all SNPs estimated to be present. In addition, regions that are missing or highly divergent in at least one accession encompass about a quarter of the reference genome .
The progress made with genome-wide association (GWA) mapping in humans during the past three years has been nothing but phenomenal , and bodes well for applying association mapping to A. thaliana. As in humans, linkage disequilibrium (LD), which is the basis for GWA studies, decays over about 10 kb, the equivalent of two average genes . That the average LD in Arabidopsis is not so different from that in humans might seem surprising, given the selfing nature of A. thaliana, but it reflects the fact that outcrossing is not that rare, and that this species apparently has a large effective population size. A 250 k SNP chip (containing 250,000 probes), corresponding to approximately one SNP very 480 bp, has been produced, and should predict some 90% of all non-singleton SNPs . A collection of over 6,000 A. thaliana accessions, both from stock centers and recent collections (for example ) has been assembled, and a subset of 1,200 genetically diverse strains will be interrogated with the 250 k SNP chip , providing a fantastic resource for GWA studies in this species.
It is becoming increasingly clear that it is inappropriate to think about 'the' genome of a species, even though this is what the initial sequencing papers stated in their titles just a few years ago (as in "Initial sequencing and analysis of the human genome" and "The sequence of the human genome") [27,28]. The previous emphasis on relatively minor changes between individuals, such as SNPs and small indels, was largely due to the fact that sequence variation had overwhelmingly been studied by PCR-based methods or hybridization to known sequences. It is now known that A. thaliana accessions can vary in hundreds of genes [21,29], and similar findings have emerged for other species, including humans (for example [30,31]). Of particular importance is the observation that some genes with fundamental effects on life-history traits such as flowering are not even functional in the A. thaliana Col-0 reference accession , and thus could not have been discovered on the basis of the first genome sequence alone.
The 250 k SNP genotyping effort discussed above is an important step towards identifying haplotype blocks associated with specific trait variants, but it has several limitations. First, the initial SNP discovery phase had considerable, technology-inherent shortcomings, and only a minority of all SNPs was detected . Second, these SNPs were defined in a relatively small initial sample that probably captures only a fraction of species-wide diversity. Genotyping with SNPs common in the global population will provide little information on new alleles that have arisen on the background of older haplotypes, which would be particularly relevant for studies of local populations. Third, although the impact of structural variation is unknown, it might have dramatic consequences on phenotypic diversity.
Together with partners from around the world, we have initiated a project with the goal of describing the whole-genome sequence variation in 1,001 accessions of A. thaliana . The current technological revolution in sequencing means that it is now feasible and inexpensive to sequence large numbers of genomes. Indeed, a 1000 Genomes Project for humans was announced in January 2008 , and the first results of this initiative are very encouraging [34,35]. It builds, in a manner similar to the A. thaliana project, on previous HapMap information, but because of the greater complexity and repetitiveness of human genomes, much of the initial effort for the human project will go towards comparing the feasibility of different approaches. In contrast, even short reads of the A. thaliana sequence, such as those produced by the first generation of Illumina's Genome Analyzer instrument, have already been proved to support not only the discovery of SNPs, but also of short to medium-size indels, including the detection of sequences not present in the reference genome .
We are proposing a hierarchical strategy to sequence the species-wide genome of A. thaliana. The first aspect of this approach is to make use of different technologies and different depths of sequencing coverage. A small number of genome sequences that approach the quality of the original Col-0 reference will be generated by exploiting mostly technologies such as Roche's 454 platform, which generates longer reads, in combination with libraries of different insert sizes, allowing long-range assembly. A much larger number of genomes will be sequenced with a less expensive technology such as Illumina's Genome Analyzer or Applied Biosystems' SOLiD and with only a single type of clone library. For this set of accessions, local haplotype similarity will be exploited in combination with information from the reference genomes to deduce the complete sequence, using methods similar those employed in inbred strains of mice . The power of this approach is in the large number of accessions that can be sequenced. For example, even if a particular haplotype is only present at 1% frequency, and each of the 1,001 strains is only sequenced at 8× coverage, there would still be on average 80 reads for each site in this haplotype.
The second aspect of the hierarchical approach will be the sampling of ten individuals from ten populations each in ten geographic regions throughout Eurasia, plus at least one North African accession (10 × 10 × 10 + 1) (see Figure Figure1a).1a). We expect individuals from the same region to show more extensive haplotype sharing than is observed in worldwide samples [4,24], which will be advantageous for the imputation strategy discussed above. An argument that might be raised against this approach is the strong population structure it entails, but we note that it is probably impossible to sample accessions in a manner that avoids population structure completely, and that our strategy will allow us to address questions of local adaptation, which are of great interest to evolutionary scientists. The output of the 1001 Genomes project will be a generalized genome sequence that encompasses every A. thaliana accession analysed as a special case. It will comprise a mosaic of variable haplotypes such that every genome can be aligned completely against it.
It is instructive to compare our proposal with the 1000 Genomes effort for humans  and the Drosophila Genetic Reference Panel projects . Because A. thaliana accessions are inbred with effectively constant genomes, and can be readily distributed as seeds, the genome sequence data we generate can be used directly in association mapping; of particular importance, the causative mutations will be observed in most cases. In contrast, the human population is not made up of highly inbred individuals, and the genetic variation discovered in 1000 humans is only a first step, yielding a deep catalog of genetic variation that allows one to infer indirectly much of the genome sequence in the samples used in association studies . The A. thaliana 1001 Genomes project is relatively simple compared with its bigger human cousin, and much more affordable because A. thaliana genomes are about 20 times smaller than human genomes (40 times, if one counts both homologs in the outbred genomes of our species). Consequently, the powerful arguments that justified funding the human effort are even more persuasive in the case of A. thaliana. Indeed, the reasoning for the Drosophila Genetic Reference Panel  spearheaded by Trudy Mackay is very similar to that advanced for the A. thaliana project. Important differences are, however, that Drosophila melanogaster does not self-fertilize. Inbred lines therefore have to be derived by repeated brother-sister matings, and although they capture variation present in nature, wild individuals are genetically more complex. Moreover, the initial Drosophila 192 lines, which are the focus of this project, were collected from a single locale, in contrast to the much wider sampling for both the human and the A. thaliana projects.
Some of the A. thaliana genomes will be immediately useful, as they are from parents of recombinant inbred line populations, a widely used resource for QTL mapping in A. thaliana . The genome sequences will provide information on potential functional polymorphisms responsible for the identified QTL.
The main motivation for the 1001 Genomes project is, however, to enable GWA studies in this species. The seeds from the 1,001 accessions will be freely available from the Arabidopsis stock centers , and each accession can be grown and phenotyped by scientists from all over the world, in as many environments as desired. Importantly, because an unlimited supply of genetically identical individuals will be available for each accession, even subtle phenotypes and ones that are highly sensitive to the microenvironment, which is often difficult to control, can be measured with high confidence. The phenotypes will include morphological analyses, such as plant stature, growth and flowering; investigations of plant content, such as metabolites and ions; responses to the abiotic environment, such as resistance to drought or salt stress; or resistance to disease caused by a host of prokaryotic and eukaryotic pathogens, from microbes to insects and nematodes. In the last case, a particularly exciting prospect is the ability to identify plant genes that mediate the effects of individual pathogen proteins, which are normally delivered as a complex mix to the plant, as is being done in the Effectoromics project, which has the aim of "understanding host plant susceptibility and resistance by indexing and deploying obligate pathogen effectors" . The value of being able to correlate many different phenotypes, including genome-wide phenotypes, has already been beautifully demonstrated for the Drosophila Genetic Reference Panel , and we expect similar dividends for the A. thaliana project.
We envisage that ultimately there will be web-based tools for GWA scans to identify candidate polymorphisms affecting these phenotypes in the 1,001 accessions. As part of the Arabidopsis 2010 Project, the US National Science Foundation is already supporting the development of web resources that will help the wider community to exploit such sequence data . It goes without saying that one needs to employ appropriate statistical methods to control for population structure caused by the hierarchical choice of accessions, which might otherwise produce false-positive associations.
A potential shortcoming of GWA scans is that some alleles responsible for interesting traits are strongly partitioned between different populations. They are in strong LD with many physically unlinked loci and thus difficult to pinpoint. A powerful approach to circumvent such problems of population structure is the generation of experimental populations in which members of different populations are intercrossed in a systematic way. Such a strategy, dubbed nested association mapping (NAM), has been developed for maize , and similar designs are being used in mice [44,45]. Corresponding efforts are under way for A. thaliana as well . As part of the 1001 Genomes Project, the parental accessions in these lines are already being sequenced, which will enable the reconstruction of complete haplotype maps in the hundreds of derived intercrossed lines, which need to be characterized at only a relatively modest number of informative SNPs. Association scans with this material will provide an extremely useful complement to conventional GWA. In future phenotyping projects, it might be advisable to split efforts between wild accessions and the intercrossed lines.
This leaves the question: why 1,001 genomes, and not 101 or 10,001? As with the human 1000 Genomes project, 1,001 is obviously an arbitrarily chosen number, to capture the imagination of our colleagues (and of the funding agencies). Some might argue that rather than sequencing 1,001 A. thaliana accessions, one should sequence, say, 200 A. thaliana strains and 200 rice strains. Our answer is that we see the A. thaliana 1001 Genomes project only as a first feasibility study, and that we are fully expecting similar projects for rice and other crops to follow soon. The dawn of a new era of plant genetics is truly upon us.
We thank our many colleagues around the world, including Joe Ecker (Salk Institute), Wolf Frommer and Len Penacchio (JGI and JBEI), Christian Hardtke (Lausanne), Jonathan Jones (Sainsbury Laboratory), Todd Michael (Waksman Institute), and Magnus Nordborg (USC/GMI), for contributing to the 1001 Genomes vision. Arabidopsis thaliana sequencing efforts in our labs are supported by the BBSRC (RM), BMBF (ERA-PG ARABRAS and GABI-GNADE), a Gottfried Wilhelm Leibniz Award (DFG) and the Max Planck Society (DW).