Main

The plant and animal kingdoms evolved independently from unicellular eukaryotes and represent highly contrasting life forms. The genome sequences of C. elegans1 and Drosophila2 reveal that metazoans share a great deal of genetic information required for developmental and physiological processes, but these genome sequences represent a limited survey of multicellular organisms. Flowering plants have unique organizational and physiological properties in addition to ancestral features conserved between plants and animals. The genome sequence of a plant provides a means for understanding the genetic basis of differences between plants and other eukaryotes, and provides the foundation for detailed functional characterization of plant genes.

Arabidopsis thaliana has many advantages for genome analysis, including a short generation time, small size, large number of offspring, and a relatively small nuclear genome. These advantages promoted the growth of a scientific community that has investigated the biological processes of Arabidopsis and has characterized many genes3. To support these activities, an international collaboration (the Arabidopsis Genome Initiative, AGI) began sequencing the genome in 1996. The sequences of chromosomes 2 and 4 have been reported4,5, and the accompanying Letters describe the sequences of chromosomes 1 (ref. 6), 3 (ref. 7) and 5 (ref. 8).

Here we report analysis of the completed Arabidopsis genome sequence, including annotation of predicted genes and assignment of functional categories. We also describe chromosome dynamics and architecture, the distribution of transposable elements and other repeats, the extent of lateral gene transfer from organelles, and the comparison of the genome sequence and structure to that of other Arabidopsis accessions (distinctive lines maintained by single-seed descent) and plant species. This report is the summation of work by experts interested in many biological processes selected to illuminate plant-specific functions including defence, photomorphogenesis, gene regulation, development, metabolism, transport and DNA repair.

The identification of many new members of receptor families, cellular components for plant-specific functions, genes of bacterial origin whose functions are now integrated with typical eukaryotic components, independent evolution of several families of transcription factors, and suggestions of as yet uncharacterized metabolic pathways are a few more highlights of this work. The implications of these discoveries are not only relevant for plant biologists, but will also affect agricultural science, evolutionary biology, bioinformatics, combinatorial chemistry, functional and comparative genomics, and molecular medicine.

The Arabidopsis Genome Initiative

Three groups contributed to the work reported here. The Genome Sequencing groups, arranged here in order of sequence contribution, sequenced and annotated assigned chromosomal regions. The Genome Analysis group carried out the analyses described. The Contributing Authors interpreted the genome analyses, incorporating other data and analyses, with respect to selected biological topics.

Genome Sequencing Groups

Samir Kaul, Hean L. Koo, Jennifer Jenkins, Michael Rizzo, Timothy Rooney, Luke J. Tallon, Tamara Feldblyum, William Nierman, Maria-Ines Benito, M-I Xiaoying Lin, Christopher D. Town, J. Craig Venter, Claire M. Fraser, Satoshi Tabata, Yasukazu Nakamura, Takakazu Kaneko, Shusei Sato, Erika Asamizu, Tomohiko Kato, Hirokazu Kotani, Shigemi Sasamoto, Joseph R. Ecker, Athanasios Theologis,Nancy A. Federspiel, Curtis J. Palm, Brian I. Osborne, Paul Shinn, Aaron B. Conway, Valentina S. Vysotskaia, Ken Dewar, Lane Conn, Catherine A. Lenz, Christopher J. Kim, Nancy F. Hansen, Shirley X. Liu, Eugen Buehler, Hootan Altafi, Hitomi Sakano,Patrick Dunn, Bao Lam, Paul K. Pham, Qimin Chao, Michelle Nguyen, Guixia Yu, Huaming Chen, Audrey Southwick, Jeong Mi Lee, Molly Miranda, Mitsue J. Toriumi, Ronald W. Davis.

European Union Chromosome 4 and 5 Sequencing Consortium

R. Wambutt, G. Murphy, A. Düsterhöft, W. Stiekema, T. Pohl, K.-D. Entian,N. Terryn, G. Volckaert.

European Chromosome 3 Sequencing Consortium

M. Salanoubat, N. Choisne, M. Rieger, W. Ansorge, M. Unseld, B. Fartmann, G. Valle, F. Artiguenave, J. Weissenbach, F. Quetier.

The Cold Spring Harbor and Washington University Genome SequencingCenter Consortium

Richard K. Wilson, Melissa de la Bastide, M. Sekhon, Emily Huang, Lori Spiegel, Lidia Gnoj, K. Pepin, J. Murray, D. Johnson, Kristina Habermann, Neilay Dedhia, Larry Parnell, Raymond Preston, L. Hillier, Ellson Chen, M. Marra, Robert Martienssen, W. Richard McCombie.

Genome Analysis Group

Klaus Mayer, Owen White, Michael Bevan, Kai Lemcke, Todd H. Creasy, Cord Bielke, Brian Haas, Dirk Haase, Rama Maiti, Stephen Rudd, Jeremy Peterson, Heiko Schoof, Dimitrij Frishman, Burkhard Morgenstern, Paulo Zaccaria, Maria Ermolaeva, Mihaela Pertea, John Quackenbush, Natalia Volfovsky, Dongying Wu, Todd M. Lowe, Steven L. Salzberg, Hans-Werner Mewes.

CONTRIBUTING AUTHORS

Comparative analysis of the genomes of A. thaliana accessions

S. Rounsley, D. Bush, S. Subramaniam, I. Levin, S. Norris.

Comparative analysis of the genomes of A. thaliana and othergenera

R. Schmidt, A. Acarkan, I. Bancroft.

Integration of the three genomes in the plant cell: the extent of protein and nucleic acid traffic between nucleus, plastids and mitochondria

F. Quetier, A. Brennicke, J. A. Eisen.

Transposable elements

T. Bureau, B.-A. Legault, Q.-H. Le, N. Agrawal, Z. Yu, R. Martienssen.

rDNA, telomeres and centromeres

G. P. Copenhaver, S. Luo, C. S. Pikaard, D. Preuss.

Membrane transport

I. T. Paulsen, M. Sussman.

DNA repair and recombination

A. B. Britt, J. A. Eisen.

Gene regulation

D. A. Selinger, R. Pandey, D. W. Mount, V. L. Chandler, R. A. Jorgensen, C. Pikaard.

Cellular organization

G. Juergens.

Development

E. M. Meyerowitz.

Signal transduction

J. R. Ecker, A. Theologis.

Recognition of and response to pathogens

J. Dangl, J. D. G. Jones.

Photomorphogenesis and photosynthesis

M. Chen, J. Chory.

Metabolism

C. Somerville.

Overview of sequencing strategy

We used large-insert bacterial artificial chromosome (BAC), phage (P1) and transformation-competent artificial chromosome (TAC) libraries9,10,11,12 as the primary substrates for sequencing. Early stages of genome sequencing used 79 cosmid clones. Physical maps of the genome of accession Columbia were assembled by restriction fragment ‘fingerprint’ analysis of BAC clones13, by hybridization14 or polymerase chain reaction (PCR)15 of sequence-tagged sites and by hybridization and Southern blotting16. The resulting maps were integrated (http://nucleus/cshl.org/arabmaps/) with the genetic map and provided a foundation for assembling sets of contigs into sequence-ready tiling paths. End sequence (http://www.tigr.org/tdb/at/abe/bac_end_search.html) of 47,788 BAC clones was used to extend contigs from BACS anchored by marker content and to integrate contigs.

Ten contigs representing the chromosome arms and centromeric heterochromatin were assembled from 1,569 BAC, TAC, cosmid and P1 clones (average insert size 100 kilobases (kb)). Twenty-two PCR products were amplified directly from genomic DNA and sequenced to link regions not covered by cloned DNA or to optimize the minimal tiling path. Telomere sequence was obtained from specific yeast artificial chromosome (YAC) and phage clones, and from inverse polymerase chain reaction (IPCR) products derived from genomic DNA. Clone fingerprints, together with BAC end sequences, were generally adequate for selection of clones for sequencing over most of the genome. In the centromeric regions, these physical mapping methods were supplemented with genetic mapping to identify contig positions and orientation17.

Selected clones were sequenced on both strands and assembled using standard techniques. Comparison of independently derived sequence of overlapping regions and independent reassembly sequenced clones revealed accuracy rates between 99.99 and 99.999%. Over half of the sequence differences were between genomic and BAC clone sequence. All available sequenced genetic markers were integrated into sequence assemblies to verify sequence contigs4,5,6,7,8. The total length of sequenced regions, which extend from either the telomeres or ribosomal DNA repeats to the 180-base-pair (bp) centromeric repeats, is 115,409,949 bp (Table 1). Estimates of the unsequenced centromeric and rDNA repeat regions measure roughly 10 megabases (Mb), yielding a genome size of about 125 Mb, in the range of the 50–150 Mb haploid content estimated by different methods18. In general, features such as gene density, expression levels and repeat distribution are very consistent across the five chromosomes (Fig. 1), and these are described in detail in reports on individual chromosomes4,5,6,7,8 and in the analysis of centromere, telomere and rDNA sequences.

Table 1 Summary statistics of the Arabidopsis genome
Figure 1: Representation of the Arabidopsis chromosomes.
figure 1

Each chromosome is represented as a coloured bar. Sequenced portions are red, telomeric and centromeric regions are light blue, heterochromatic knobs are shown black and the rDNA repeat regions are magenta. The unsequenced telomeres 2N and 4N are depicted with dashed lines. Telomeres are not drawn to scale. Images of DAPI-stained chromosomes were kindly supplied by P. Fransz. The frequency of features was given pseudo-colour assignments, from red (high density) to deep blue (low density). Gene density (‘Genes’) ranged from 38 per 100 kb to 1 gene per 100 kb; expressed sequence tag matches (‘ESTs’) ranged from more than 200 per 100 kb to 1 per 100 kb. Transposable element densities (‘TEs’) ranged from 33 per 100 kb to 1 per 100 kb. Mitochondrial and chloroplast insertions (‘MT/CP’) were assigned black and green tick marks, respectively. Transfer RNAs and small nucleolar RNAs (‘RNAs’) were assigned black and red ticks marks, respectively.

We used tRNAscan-SE 1.21 (ref. 19) and manual inspection to identify 589 cytoplasmic transfer RNAs, 27 organelle-derived tRNAs and 13 pseudogenes—more than in any other genome sequenced to date. All 46 tRNA families needed to decode all possible 61 codons were found, defining the completeness of the functional set. Several highly amplified families of tRNAs were found on the same strand6; excluding these, each amino acid is decoded by 10–41 tRNAs.

The spliceosomal RNAs (U1, U2, U4, U5, U6) have all been experimentally identified in Arabidopsis. The previously identified sequences for all RNAs were found in the genome, except for U5 where the most similar counterpart was 92% identical. Between 10 and 16 copies of each small nuclear RNA (snRNA) were found across all chromosomes, dispersed as singletons or in small groups.

The small nucleolar RNAs (snoRNAs) consist of two subfamilies, the C/D box snoRNAs, which includes 36 Arabidopsis genes, and the H/ACA box snoRNAs, for which no members have been identified in Arabidopsis. U3 is the most numerous of the C/D box snoRNAs, with eight copies found in the genome. We identified forty-five additional C/D box snoRNAs using software (www.rna.wustl.edu/snoRNAdb/) that detects snoRNAs that guide ribose methylation of ribosomal RNA.

A combination of algorithms, all optimized with parameters based on known Arabidopsis gene structures, was used to define gene structure. We used similarities to known protein and expressed sequence tag (EST) sequence to refine gene models. Eighty per cent of the gene structures predicted by the three centres involved were completely consistent, 93% of ESTs matched gene models, and less than 1% of ESTs matched predicted non-coding regions, indicating that most potential genes were identified. The sensitivity and selectivity of the gene prediction software used in this report has been comprehensively and independently assessed20.

The 25,498 genes predicted (Table 1) is the largest gene set published to date: C. elegans1 has 19,099 genes and Drosophila2 13,601 genes. Arabidopsis and C. elegans have similar gene density, whereas Drosophila has a lower gene density; Arabidopsis also has a significantly greater extent of tandem gene duplications and segmental duplications, which may account for its larger gene set.

The rDNA repeat regions on chromosomes 2 and 4 were not sequenced because of their known repetitive structure and content. The centromeric regions are not completely sequenced owing to large blocks of monotonic repeats such as 5S rDNA and 180-bp repeats. The sequence continues to be extended further into centromeric and other regions of complex sequence.

Characterization of the coding regions

To assess the similarities and differences of the Arabidopsis gene complement compared with other sequenced eukaryotic genomes, we assigned functional categories to the complete set of Arabidopsis genes. For chromosome 4 genes and the yeast genome, predicted functions were previously manually assigned5,21. All other predicted proteins were automatically assigned to these functional categories22, assuming that conserved sequences reflect common functional relationships.

The functions of 69% of the genes were classified according to sequence similarity to proteins of known function in all organisms; only 9% of the genes have been characterized experimentally (Fig. 2a ). Generally similar proportions of gene products were predicted to be targeted to the secretory pathway and mitochondria in Arabidopsis and yeast, and up to 14% of the gene products are likely to be targeted to the chloroplast (Table 1). The significant proportion of genes with predicted functions involved in metabolism, gene regulation and defence is consistent with previous analyses5. Roughly 30% of the 25,498 predicted gene products, (Fig. 2a), comprising both plant-specific proteins and proteins with similarity to genes of unknown function from other organisms, could not be assigned to functional categories.

Figure 2: Functional analysis of Arabidopsis genes.
figure 2

a, Proportion of predicted Arabidopsis genes in different functional categories. b, Comparison of functional categories between organisms. Subsets of the Arabidopsis proteome containing all proteins that fall into a common functional class were assembled. Each subset was searched against the complete set of translations from Escherichia coli, Synechocystis sp. PCC6803, Saccharomyces cerevisae, Drosophila , C. elegans and a Homo sapiens non-redundant protein database. The percentage of Arabidopsis proteins in a particular subset that had a BLASTP match with E ≤ 10-30 to the respective reference genome is shown. This reflects the measure of sequence conservation of proteins within this particular functional category between Arabidopsis and the respective reference genome. y axis, 0.1 = 10%.

To compare the functional catagories in more detail, we compared data from the complete genomes of Escherichia coli23, Synechocystis sp.24, Saccharomyces cerevisiae21, C. elegans1 and Drosophila2, and a non-redundant protein set of Homo sapiens, with the Arabidopsis genome data (Fig. 2b), using a stringent BLASTP threshold value of E < 10-30. The proportion of Arabidopsis proteins having related counterparts in eukaryotic genomes varies by a factor of 2 to 3 depending on the functional category. Only 8–23% of Arabidopsis proteins involved in transcription have related genes in other eukaryotic genomes, reflecting the independent evolution of many plant transcription factors. In contrast, 48–60% of genes involved in protein synthesis have counterparts in the other eukaryotic genomes, reflecting highly conserved gene functions. The relatively high proportion of matches between Arabidopsis and bacterial proteins in the categories ‘metabolism’ and ‘energy’ reflects both the acquisition of bacterial genes from the ancestor of the plastid and high conservation of sequences across all species. Finally, a comparison between unicellular and multicellular eukaryotes indicates that Arabidopsis genes involved in cellular communication and signal transduction have more counterparts in multicellular eukaryotes than in yeast, reflecting the need for sets of genes for communication in multicellular organisms.

Pronounced redundancy in the Arabidopsis genome is evident in segmental duplications and tandem arrays, and many other genes with high levels of sequence conservation are also scattered over the genome. Sequence similarity exceeding a BLASTP value E < 10-20 and extending over at least 80% of the protein length were used as parameters to identify protein families (Table 2). A total of 11,601 protein types were identified. Thirty-five per cent of the predicted proteins are unique in the genome, and the proportion of proteins belonging to families of more than five members is substantially higher in Arabidopsis (37.4%) than in Drosophila (12.1%) or C. elegans (24.0%). The absolute number of Arabidopsis gene families and singletons (types) is in the same range as the other multicellular eukaryotes, indicating that a proteome of 11,000–15,000 types is sufficient for a wide diversity of multicellular life. The proportion of gene families with more than two members is considerably more pronounced in Arabidopsis than in other eukaryotes ( Fig. 3). As segmental duplication is responsible for 6,303 gene duplications (see below), the extent of tandem gene duplications accounts for a significant proportion of the increased family size. These features of the Arabidopsis , and presumably other plant genomes, may indicate more relaxed constraints on genome size in plants, or a more prominent role of unequal crossing over to generate new gene copies.

Table 2 Proportion of genes in different organisms present as either singletons or in paralogous families
Figure 3: Distribution of tandemly repeated gene arrays in the Arabidopsis genome.
figure 3

Tandemly repeated gene arrays were identified using the BLASTP program with a threshold of E < 10-20. One unrelated gene among cluster members was tolerated. The histogram gives the number of clusters in the genome containing 2 to n similar gene units in tandem.

Conserved protein domains revealed more informative differences through INTERPRO25 analysis of the predicted gene products from Arabidopsis, S. cerevisiae, C. elegans and Drosophila . Statistically over-represented domains, and those that are absent from the Arabidopsis genome, indicate domains that may have been gained or lost during the evolution of plants (Supplementary Information Table 1). Proteins containing the Pro-Pro-Arg repeat, which is involved in RNA stabilization and RNA processing, are over-represented as compared to yeast, fly and worm; 400 proteins containing this signature were detected in Arabidopsis compared with only 10 in total in yeast, Drosophila and C. elegans. Protein kinases and associated domains, 169 proteins containing a disease resistance protein signature, and the Toll/IL-1R (TIR) domain, a component of pathogen recognition molecules26, are also relatively abundant. This suggests that pathways transducing signals in response to pathogens and diverse environmental cues are more abundant in plants than in other organisms.

The RING zinc finger domain is relatively over-represented in Arabidopsis compared with yeast, Drosophila and C. elegans, whereas the F-box domain is over-represented as compared with yeast and Drosophila only. These domains are involved in targeting proteins to the proteasome27 and ubiquitinylation28 pathways of protein degradation, respectively. In plants many processes such as hormone and defence responses, light signalling, and circadian rhythms and pattern formation use F-box function to direct negative regulators to the ubiquitin degradation pathway. This mode of regulation appears to be more prevalent in plants and may account for a higher representation of the F box than in Drosophila and for the over-representation of the ubiquitin domain in the Arabidopsis genome. RING finger domain proteins in general have a role in ubiquitin protein ligases, indicating that proteasome-mediated degradation is a more widespread mode of regulation in plants than in other kingdoms.

Most functions identified by protein domains are conserved in similar proportions in the Arabidopsis, S. cerevisiae, Drosophila and C. elegans genomes, pointing to many ubiquitous eukaryotic pathways. These are illustrated by comparing the list of human disease genes29 to the complete Arabidopsis gene set using BLASTP. Out of 289 human disease genes, 139 (48%) had hits in Arabidopsis using a BLASTP threshold E < 10-10. Sixty-nine (24%) exceeded an E < 10-40 threshold, and 26 (9.3%) had scores better than E < 10-100 (Table 3). There are at least 17 human disease genes more similar to Arabidopsis genes than yeast, Drosophila or C. elegans genes (Table 3).

Table 3 Arabidopsis genes with similarities to human disease genes

This analysis shows that, although numerous families of proteins are shared between all eukaryotes, plants contain roughly 150 unique protein families. These include transcription factors, structural proteins, enzymes and proteins of unknown function. Members of the families of genes common to all eukaryotes have undergone substantial increases or decreases in their size in Arabidopsis . Finally, the transfer of a relatively small number of cyanobacteria-related genes from a putative endosymbiotic ancestor of the plastid has added to the diversity of protein structures found in plants.

Genome organization and duplication

The Arabidopsis genome sequence provides a complete view of chromosomal organization and clues to its evolutionary history. Gene families organized in tandem arrays of two or more units have been described in C. elegans 1 and Drosophila2. Analysis of the Arabidopsis genome revealed 1,528 tandem arrays containing 4,140 individual genes, with arrays ranging up to 23 adjacent members (Fig.3 ). Thus 17% of all genes of Arabidopsis are arranged in tandem arrays.

Large segmental duplications were identified either by directly aligning chromosomal sequences or by aligning proteins and searching for tracts of conserved gene order. All five chromosomes were aligned to each other in both orientations using MUMmer30, and the results were filtered to identify all segments at least 1,000 bp in length with at least 50% identity (Supplementary Information Fig. 1). These revealed 24 large duplicated segments of 100 kb or larger, comprising 65.6 Mb or 58% of the genome. The only duplicated segment in the centromeric regions was a 375-kb segment on chromosome 4. Many duplications appear to have undergone further shuffling, such as local inversions after the duplication event.

We used TBLASTX5 to identify collinear clusters of genes residing in large duplicated chromosomal segments. The duplicated regions encompass 67.9 Mb, 60% of the genome, slightly more than was found in the DNA-based alignment (Fig. 4), and these data extend earlier findings4,5,31. The extent of sequence conservation of the duplicated genes varies greatly, with 6,303 (37%) of the 17,193 genes in the segments classified as highly conserved (E < 10-30) and a further 1,705 (10%) showing less significant similarity up to E < 10-5. The proportion of homologous genes in each duplicated segment also varies widely, between 20% and 47% for the highly conserved class of genes. In many cases, the number of copies of a gene and its counterpart differ (for example, one copy on one chromosome and multiple copies on the other; see Supplementary Information Fig. 2); this could be due to either tandem duplication or gene loss after the segmental duplication.

Figure 4: Segmentally duplicated regions in the Arabidopsis genome.
figure 4

Individual chromosomes are depicted as horizontal grey bars (with chromosome 1 at the top), centromeres are marked black. Coloured bands connect corresponding duplicated segments. Similarity between the rDNA repeats are excluded. Duplicated segments in reversed orientation are connected with twisted coloured bands. The scale is in megabases.

What does the duplication in the Arabidopsis genome tell us about the ancestry of the species? Polyploidy occurs widely in plants and is proposed to be a key factor in plant evolution32. As the majority of the Arabidopsis genome is represented in duplicated (but not triplicated) segments, it appears most likely that Arabidopsis, like maize, had a tetraploid ancestor33. A comparative sequence analysis of Arabidopsis and tomato estimated that a duplication occurred 112 Myr ago to form a tetraploid34. The degrees of conservation of the duplicated segments might be due to divergence from an ancestral autotetraploid form, or might reflect differences present in an allotetraploid ancestor. It is also possible, however, that several independent segmental duplication events took place instead of tetraploid formation and stabilization.

The diploid genetics of Arabidopsis and the extensive divergence of the duplicated segments have masked its evolutionary history. The determination of Arabidopsis gene functions must therefore be pursued with the potential for functional redundancy taken into account. The long period of time over which genome stabilization has occurred has, however, provided ample opportunity for the divergence of the functions of genes that arose from duplications.

Comparative analysis of Arabidopsis accessions

Comparing the multiple accessions of Arabidopsis allows us to identify commonly occurring changes in genome microstructure. It also enables the development of new molecular markers for genetic mapping. High rates of polymorphism between Arabidopsis accessions, including both DNA sequence and copy number of tandem arrays, are prevalent at loci involved in disease resistance35. This has been observed for other plant species, and such loci are thought to serve as templates for illegitimate recombination to create new pathogen response specificities36. We carried out a comparative analysis between 82 Mb of the genome sequence of Arabidopsis accession Columbia (Col-0) and 92.1 Mb of non-redundant low-pass (twofold redundant) sequence data of the genomic DNA of accession Landsberg erecta (L er). We identified two classes of differences between the sequences: single nucleotide polymorphisms (SNPs), and insertion–deletions (InDels). As we used high stringency criteria, our results represent a minimum estimate of numbers of polymorphisms between the two genomes.

In total, we detected 25,274 SNPs, representing an average density of 1 SNP per 3.3 kb. Transitions (A/T–G/C) represented 52.1% of the SNPs, and transversions accounted for the remainder: 17.3% for A/T–T/A, 22.7% for A/T–C/G and 7.9% for C/G–G/C. In total, we detected 14,570 InDels at an average spacing of 6.1 kb. They ranged from 2 bp to over 38 kilobase-pairs, although 95% were smaller than 50 bp. Only 10% of the InDels were co-located with simple sequence repeats identified with the program Sputnik. An analysis of 416 relative insertions greater than 250 bp in Col-0 showed that 30% matched transposon-related proteins, indicating that a substantial proportion of the large InDels are the result of transposon insertion or excision. Many InDels contained entire active genes not related to transposons. Half of such genes absent from corresponding positions in the Col-0 sequence were found elsewhere on the genome of Ler. This indicates that genes have been transferred to new genomic locations.

Gene structures are often affected by small InDels and SNPs. The positions of SNPs and InDels were mapped relative to 87,427 exons and 70,379 introns annotated in the Col-0 sequence. SNPs were found in exons, introns and intergenic regions at frequencies of 1 SNP per 3.1, 2.2 and 3.5 kb, respectively. The frequencies for InDels were 1 per 9.3, 3.1 and 4.3 kb, respectively. Polymorphisms were detected in 7% of exons, and alter the spliced sequences of 25% of the predicted genes. For InDels in exons, insertion lengths divisible by three are prevalent for small insertions (< 50 bp), indicating that many proteins can withstand small insertions or deletions of amino acids without loss of function.

Our analyses show that sequence polymorphisms between accessions of Arabidopsis are common, and that they occur in both coding and non-coding regions. We found evidence for the relocation of genes in the genome, and for changes in the complement of transposable elements. The data presented here are available at http://www.arabidopsis.org/cereon/.

Comparison of Arabidopsis and other plant genera

Comparative genetic mapping can reveal extensive conservation of genome organization between closely related species37,38. The comparative analysis of plant genome microstructure reveals much about the evolution of plant genomes and provides unprecedented opportunities for crop improvement by establishing the detailed structures of, and relationships between, the genomes of crops and Arabidopsis.

The lineages leading to Arabidopsis and Capsella rubella (shepherd' s purse) diverged between 6.2 and 9.8 Myr ago, and the gene content and genome organization of C. rubella is very similar to that of Arabidopsis39, including the large-scale duplications. Alignment of Arabidopsis complementary DNA and EST sequences with genomic DNA sequences of Arabidopsis and C. rubella showed conservation of exon length and intron positions. Coding sequences predicted from these alignments differed from the annotated Arabidopsis gene sequences in two out of five cases.

The ancestral lineages of Arabidopsis and the Brassica (cabbage and mustard) genera diverged 12.2–19.2 Myr ago40. Brassica genes show a high level of nucleotide conservation with their Arabidopsis orthologues, typically more than 85% in coding regions40. The structure of Brassica genomes resembles that of Arabidopsis, but with extensive triplication and rearrangement41, and extensive divergence of microstructure (Supplementary Information Fig. 3). The divergence between the genomes of Arabidopsis and Brassica oleracea is in striking contrast to that observed between Arabidopsis and C. rubella, although the time since divergence is only twofold greater. This accelerated rate of change in triplicated segments of the genome of B. oleracea indicates that polyploidy fosters rapid chromosomal evolution.

The Arabidopsis and tomato lineages diverged roughly 150 Myr ago, and comparative sequence analysis of segments of their genomes has revealed complex relationships34. Four regions of the Arabidopsis genome are related to each other and to one region in the tomato genome, suggesting that two rounds of duplication may have occurred in the Arabidopsis lineage. The extensive duplication described here supports the proposal that the more recent of these duplications, estimated to have occurred 112 Myr ago, was the result of a polyploidization event. The lineages of Arabidopsis and rice diverged 200 Myr ago42. Three regions of the genome of Arabidopsis were related to each other and to one region in the rice genome, providing further evidence for multiple duplication events43,44.

The frequent occurrence of tandem gene duplications and the apparent deletion of single genes, or small groups of adjacent genes, from duplicated regions suggests that unequal crossing over may be a key mechanism affecting the evolution of plant genome microstructure. However, the segmental inversions and gene translocations in the genomes of both rice and B. oleracea that are not found in Arabidopsis indicate that additional mechanisms may be involved40.

Integration of the three genomes in the plant cell

The three genomes in the plant cell—those of the nucleus, the plastids (chloroplasts) and the mitochondria—differ markedly in gene number, organization and stability. Plastid genes are densely packed in an order highly conserved in all plants45, whereas mitochondrial genes46 are widely dispersed and subjected to extensive recombination.

Organellar genomes are remnants of independent organisms—plastids are derived from the cyanobacterial lineage and mitochondria from the α-Proteobacteria. The remaining genes in plastids include those that encode subunits of the photosystem and the electron transport chain, whereas the genes in mitochondria encode essential subunits of the respiratory chain. Both organelles contain sets of specific membrane proteins that, together with housekeeping proteins, account for 61% of the genes in the chloroplast and 88 % in the mitochondrion (Table 4). The balances are involved in transcription and translation.

Table 4 General features of genes encoded by the three genomes in Arabidopsis

The number of proteins encoded in the nucleus likely to be found in organelles was predicted using default settings on TargetP (Table 1). Many nuclear gene products that are targeted to either (or both) organelles were originally encoded in the organelle genomes and were transferred to the nuclear genome during evolutionary history. A large number also appear to be of eukaryotic origin, with functions such as protein import components, which were probably not required by the free-living ancestors of the endosymbionts.

To identify nuclear genes of possible organellar ancestry, we compared all predicted Arabidopsis proteins to all proteins from completed genomes including those from plastids and mitochondria (Supplementary Information Table 2). This search identified proteins encoded by the Arabidopsis nuclear genome that are most similar to proteins encoded by other species’ organelle genomes (14 mitochondrial and 44 plastid). These represent organelle-to-nuclear gene transfers that have occurred sometime after the divergence of the organelle-containing lineages47. There is a great excess of nuclear encoded proteins most similar to proteins from the cyanobacteria Synechocystis (Supplementary Information Fig. 4; 806 Arabidopsis predicted proteins matching 404 different Synechocystis proteins, providing further evidence of a genome duplication). These 806 Arabidopsis predicted proteins, and many others of greatly diverse function, are possibly of plastid descent. Through searches against proteins from other cyanobacteria (with incompletely sequenced genomes), we identified 69 additional genes of possibly plastid descent. Only 25% of these putatively plastid-derived proteins displayed a target peptide predicted by TargetP, indicating potential cytoplasmic functions for most of these genes.

The difference between predicted plastid-targeted and predicted plastid-derived genes indicates that there is a probable overestimation by ab initio targeting prediction methods and a lack of resolution with respect to destination organelles, the possible extensive divergence of some endosymbiont-derived genes in the nuclear genome, the co-opting of nuclear genes for targeting to organelles, and cytoplasmic functions for cyanobacteria-derived proteins. Clearly more refined tools and extensive experimentation is required to catalogue plastid proteins.

The transfer of genes between genomes still continues (Supplementary Information Table 3). Plastid DNA insertions in the nucleus (17 insertions totalling 11 kb) contain full-length genes encoding proteins or tRNAs, fragments of genes and an intron as well as intergenic regions. Subsequent reshuffling in the nucleus is illustrated by the atpH gene, which was originally transferred completely, but is now in two pieces separated by 2 kb. The 13 small mitochondrial DNA insertions total 7 kb in addition to the large insertion close to the centromere of chromosome 2 (ref. 3). The high level of recombination in the mitochondrial genome may account for these events.

Transposable elements

Transposons, which were originally identified in maize by Barbara McClintock, have been found in all eukaryotes and prokaryotes. A subset of transposons replicate through an RNA intermediate (class I), whereas others move directly through a DNA form (class II). Transposons are further classified by similarity either between their mobility genes or between their terminal and/or internal motifs, as well as by the size and sequence of their target site. Internally deleted elements can often be mobilized in trans by fully functional elements.

Transposons in Arabidopsis account for at least 10% of the genome, or about one-fifth of the intergenic DNA. The Arabidopsis genome has a wealth of class I (2,109) and II (2,203) elements, including several new groups (1,209 elements; Supplementary Information Table 4). Mobile histories for many elements were obtained by identifying regions of the genome with significant similarity to ‘empty’ target sites (RESites) thus providing high-resolution information concerning the termini and target site duplications48,49. These regions were readily detected because of the propensity of transposons to integrate into repeats and because of duplications in the genome sequence. In several cases, genes appear to have been included as ‘passengers’ in transposable units48. In some cases, shared sequence similarity, coding capacity and RESites attest to recent activity of transposable elements in the Arabidopsis genome. Only about 4% of the complete elements identified correspond to an EST, however, suggesting that most are not transcribed.

Transposable elements found in many other plant genomes are well represented in Arabidopsis, including copia- and gypsy-like long terminal repeat (LTR) retrotransposons, long interspersal nuclear elements (LINEs); short interspersed nuclear elements (SINEs), hobo/Activator /Tam3 (hAT)-like elements, CACTA-like elements and miniature inverted-repeat transposable elements (MITES). Although usually small in size, some larger Tourist-like MITEs contain open reading frames (ORFs) with similarity to the transposases of bacterial insertion sequences48. Basho and many Mutator-like elements (MULEs), first discovered in the Arabidopsis sequence, represent structurally unique transposons48,49,50. Basho elements have a target site preference for mononucleotide ‘A’ and wide distribution among plants48,51. MULEs exhibit a high level of sequence diversity and members of most groups lack long terminal inverted repeats (TIRs). Phylogenetic analysis of the Arabidopsis MURA-like transposases suggests that TIR-containing MULEs are more closely related to one another than to MULEs lacking TIRs49,52.

For many plants with large genomes, class I retrotransposons contribute most of the nucleotide content53. In the small Arabidopsis genome, class I elements are less abundant and primarily occupy the centromere. In contrast, Basho elements and class II transposons such as MITEs and MULEs predominate on the periphery of pericentromeric domains ( Fig. 5). In class II transposons, MULEs and CACTA elements are clustered near centromeres and heterochromatic knobs, whereas MITEs and hAT elements have a less pronounced bias. The distribution pattern of transposable elements observed in Arabidopsis may reflect different types of pericentromeric heterochromatin regions and may be similar to those found in animals.

Figure 5: Distribution of class I, II and Basho transposons in Arabidopsis chromosomes.
figure 5

The frequency of class I retroelements (green), class II DNA transposons (blue) and Basho elements (purple) are shown at 100-kb intervals along the five chromosomes of Arabidopsis.

Numerous centromeric satellite repeats are located between each chromosome arm and have not yet been sequenced, but are represented in part by unanchored BAC contigs (R. Martienssen and M. Marra, unpublished data). End sequence suggests that these domains contain many more class I than class II elements, consistent with the distribution reported here (K. Lemcke and R. Martienssen, unpublished data). We do not know the significance of the apparent paucity of elements in telomeric regions and in the region flanking the rDNA repeats on chromosome 4 (but not on chromosome 2).

Overall, transposon-rich regions are relatively gene-poor and have lower rates of recombination and EST matches, indicating a correlation between low gene expression, high transposon density and low recombination51. The role of transposons in genome organization and chromosome structure can now be addressed in a model organism known to undergo DNA methylation and other forms of chromatin modification thought to regulate transposition52.

rDNA, telomeres and centromeres

Nucleolar organizers (NORs) contain arrays of unit repeats encoding the 18S, 5.8S and 25S ribosomal RNA genes and are transcribed by RNA polymerase I. Together with 5S RNA, which is transcribed by RNA polymerase III, these rRNAs form the structural and catalytic cores of cytoplasmic ribosomes. In Arabidopsis, the NORs juxtapose the telomeres of chromosomes 2 and 4, and comprise uninterrupted 18S, 5.8S and 25S units all orientated on the chromosomes in the same direction54. In contrast, the 5S rRNA genes are localized to heterogeneous arrays in the centromeric regions of chromosomes 3, 4 and 5 (ref. 55; and Fig.6 ). Both NORs are roughly 3.5–4.0 megabase-pairs and comprise 350–400 highly methylated rRNA gene units, each 10 kb (ref. 54). The sequence between the euchromatic arms and NORs has been determined. Elsewhere in the genome, only one other 18S, 5.8S, 25S rRNA gene unit was identified in centromere 3. Although minor variations in sequence length and composition occur in the NOR repeats, these variants are highly clustered, supporting a model of sequence maintenance through concerted evolution55.

Figure 6: Predicted centromere composition.
figure 6

Genetically defined centromere boundaries are indicated by filled circles; fully and partially assembled BAC sequences are represented by solid and dashed black lines, respectively. Estimates of repeat sizes within the centromeres were derived from consideration of repeat copy number, physical mapping and cytogenetic assays.

Arabidopsis telomeres are composed of CCCTAAA repeats and average 2–3 kb (ref. 56). For TEL4N (telomere 4 North), consensus repeats are adjacent to the NOR; the remaining telomeres are typically separated from coding sequences by repetitive subtelomeric regions measuring less than 4 kb. Imperfect telomere-like arrays of up to 24 kb are found elsewhere in the genome, particularly near centromeres. These arrays might affect the expression of nearby genes and may have resulted from ancient rearrangements, such as inversions of the chromosome arms.

Centromere DNA mediates chromosome attachment to the meiotic and mitotic spindles and often forms dense heterochromatin. Genetic mapping of the regions that confer centromere function provided the markers necessary to precisely place BAC clones at individual centromeres17; 69 clones were targeted for sequencing, resulting in over 5 Mb of DNA sequence from the centromeric regions. The unsequenced regions of centromeres are composed primarily of long, homogeneous arrays that were characterized previously with physical57 and genetic mapping17 and contain over 3 Mb of repetitive arrays, including the 180-bp repeats and 5S rDNA51 (Fig. 6).

Arabidopsis centromeres, like those of many higher eukaryotes, contain numerous repetitive elements including retroelements, transposons, microsatellites and middle repetitive DNA17. These repeats are rare in the euchromatic arms and often most abundant in pericentromeric DNA. The repeats, affinity for DNA-binding dyes, dense methylation patterns and inhibition of homologous recombination indicate that the centromeric regions are highly heterochromatic, and such regions are generally viewed as very poor environments for gene expression. Unexpectedly, we found at least 47 expressed genes encoded in the genetically defined centromeres of Arabidopsis (http://preuss.bsd.uchicago.edu/arabidopsis.genome.html). In several cases, these genes reside on islands of unique sequence flanked by repetitive arrays, such as 180-bp or 5S rDNA repeats. Among the genes encoded in the centromeres are members of 11 of the 16 functional categories that comprise the proteome. The centromeres are not subject to recombination; consequently, genes residing in these regions probably exhibit unique patterns of molecular evolution.

The function of higher eukaryotic centromeres may be specified by proteins that bind to centromere DNA, by epigenetic modifications, or by secondary or higher order structures. A pairwise comparison of the non-repetitive portions of all five centromeres showed they share limited (1–7%) sequence similarity. Forty-one families of small, conserved centromere sequences (AtCCS, see http://preuss.bsd.uchicago.edu/arabidopsis.genome.html) are enriched in the centromeric and pericentromeric regions and differ from sequences found in the centromeres of other eukaryotes. Molecular and genetic assays will be required to determine whether these conserved motifs nucleate Arabidopsis centromere activity. Apart from the AtCCS sequences, most centromere DNA is not shared between chromosomes, complicating efforts to derive clear evolutionary relationships. In contrast, genetic and cytological assays indicate that homologous centromeres are highly conserved among Arabidopsis accessions, albeit subject to rearrangements such as inversions to form knobs5,58,59 and insertions4. Further investigation of centromere DNA promises to yield information on the evolutionary forces that act in regions of limited recombination, as well as an improved understanding of the role of DNA sequence patterns in chromosome segregation.

Membrane transport

Transporters in the plasma and intracellular membranes of Arabidopsis are responsible for the acquisition, redistribution and compartmentalization of organic nutrients and inorganic ions, as well as for the efflux of toxic compounds and metabolic end products, energy and signal transduction, and turgor generation. Previous genomic analyses of membrane transport systems in S. cerevisiae and C. elegans led to the identification of over 100 distinct families of membrane transporters60,61. We compared membrane transport processes between Arabidopsis, animals, fungi and prokaryotes, and identified over 600 predicted membrane transport systems in Arabidopsis (http://www-biology.ucsd.edu/ipaulsen/transport/), a similar number to that of C. elegans (700 transporters) and over twofold greater than either S. cerevisiae or E. coli (300 transporters).

We compared the transporter complement of Arabidopsis, C. elegans and S. cerevisiae in terms of energy coupling mechanisms (Fig. 7a). Unlike animals, which use a sodium ion P-type ATPase pump to generate an electrochemical gradient across the plasma membrane, plants and fungi use a proton P-type ATPase pump to form a large membrane potential (-250 mV)62. Consequently, plant secondary transporters are typically coupled to protons rather than to sodium63. Compared with C. elegans, Arabidopsis has a surprisingly high percentage of primary ATP-dependent transporters (12% and 21% of transporters, respectively), reflecting increased numbers of P-type ATPases involved in metal ion transport and ABC ATPases proposed to be involved in sequestering unusual metabolites and drugs in the vacuole or in other intracellular compartments. These processes may be necessary for pathogen defence and nutrient storage.

Figure 7: Comparison of the transport capabilities of Arabidopsis, C. elegans and S. cerevisiae.
figure 7

Pie charts show the percentage of transporters in each organism according to bioenergetics (a) and substrate specificity (b).

About 15% of the transporters in Arabidopsis are channel proteins, five times more than in any single-celled organism but half the number in C. elegans (Fig. 7b). Almost half of the Arabidopsis channel proteins are aquaporins, and Arabidopsis has 10-fold more Mfamily major intrinsic protein (MIP) family water channels than any other sequenced organism. This abundance emphasizes the importance of hydraulics in a wide range of plant processes, including sugar and nutrient transport into and out of the vasculature, opening of stomatal apertures, cell elongation and epinastic movements of leaves and stems. Although Arabidopsis has a diverse range of metal cation transporters, C. elegans has more, many of which function in cell–cell signalling and nerve signal transduction. Arabidopsis also possesses transporters for inorganic anions such as phosphate, sulphate, nitrate and chloride, as well as for metal cation channels that serve in signal transduction or cell homeostasis. Compared with other sequenced organisms, Arabidopsis has 10-fold more predicted peptide transporters, primarily of the proton-dependent oligopeptide transport (POT) family, emphasizing the importance of peptide transport or indicating that there is broader substrate specificity than previously realized. There are nearly 1,000 Arabidopsis genes encoding Ser/Thr protein kinases, suggesting that peptides may have an important role in plant signalling64.

Virtually no transporters for carboxylates, such as lactate and pyruvate, were identified in the Arabidopsis genome. About 12% of the transporters were predicted to be sugar transporters, mostly consisting of paralogues of the MFS family of hexose transporters. Notably, S. cerevisiae, C. elegans and most prokaryotes use APC family transporters as their principle means of amino-acid transport, but Arabidopsis appears to rely primarily on the AAAP family of amino-acid and auxin transporters. More than 10% of the transporters in Arabidopsis are homologous to drug efflux pumps; these probably represent transporters involved in the sequestration into vacuoles of xenobiotics, secondary metabolites, and breakdown products of chlorophyll.

Surprisingly, Arabidopsis has close homologues of the human ABC TAP transporters of antigenic peptides for presentation to the major histocompatability complex (MHC). In Arabidopsis, these transporters may be involved in peptide efflux, or more speculatively, in some form of cell-recognition response. Arabidopsis also has 10-fold more members of the multi-drug and toxin extrusion (MATE) family than any other sequenced organism; in bacteria, these transporters function as drug efflux pumps. Curiously, Arabidopsis has several homologues of the Drosophila RND transporter family Patched protein, which functions in segment polarity, and more than ten homologues of the Drosophila ABC family eye pigment transporters. In plants, these are presumably involved in intracellular sequestration of secondary metabolites.

DNA repair and recombination

DNA repair and recombination pathways have many functions in different species such as maintaining genomic integrity, regulating mutation rates, chromosome segregation and recombination, genetic exchange within and between populations, and immune system development. Comparing the Arabidopsis genome with other species65 indicates that Arabidopsis has a similar set of DNA repair and recombination (RAR) genes to most other eukaryotes. The pathways represented include photoreactivation, DNA ligation, non-homologous end joining, base excision repair, mismatch excision repair, nucleotide excision repair and many aspects of DNA recombination ( Supplementary Information Table 5). The Arabidopsis RAR genes include homologues of many DNA repair genes that are defective in different human diseases (for example, hereditary breast cancer and non-polyposis colon cancer, xeroderma pigmentosum and Cockayne's syndrome).

One feature that sets Arabidopsis apart from other eukaryotes is the presence of additional homologues of many RAR genes. This is seen for almost every major class of DNA repair, including recombination (four RecA), DNA ligation (four DNA ligase I), photoreactivation (one class II photolyase and five class I photolyase homologues) and nucleotide excision repair (six RPA1, two RPA2, two Rad25, three TFB1 and four Rad23). This is most striking for genes with probable roles in base excision repair. Arabidopsis encodes 16 homologues of DNA base glycosylases (enzymes that recognize abnormal DNA bases and cleave them from the sugar-phosphate backbone)—more than any other species known. This includes several homologues of each of three families of alkylation damage base glycosylases: two of the S. cerevisiae MPG; six of the E. coli TagI; and two of the E. coli AlkA. Arabidopsis also encodes three homologues of the apurinic-apyrimidinic (AP) endonuclease Xth. AP endonucleases continue the base excision repair started by glycosylases by cleaving the DNA backbone at abasic sites.

Evolutionary analysis indicates that some of the extra copies of RAR genes in Arabidopsis originated through relatively recent gene duplications—because many of the sets of genes are more closely related to each other than to their homologues in any other species. As duplication is frequently accompanied by functional divergence, the duplicate (paralogous) genes may have different repair specificities or may have evolved functions that are outside RAR functions (as is the case for two of the five class I photolyase homologues, which function as blue-light receptors). In most cases, it is not known whether the paralogous gene copies have different functions. The presence of multiple paralogues might also allow functional redundancy or a greater repair or recombination capacity.

The multiplicity of RAR genes in Arabidopsis is also partly due to the transfer of genes from the organellar genomes to the nucleus. Repair gene homologues that appear to be of chloroplast origin (Supplementary Information Tables 2 and 5) include the recombination proteins RecA, RecG and SMS, two class I photolyase homologues, Fpg, two MutS2 proteins, and the transcription-repair coupling factor Mfd. Two of these (RecA and Fpg) are involved in RAR functions in the plastid, suggesting that the others may be as well. The finding of an Mfd orthologue of cyanobacterial descent is surprising. In E. coli, Mfd couples nucleotide excision repair carried out by UvrABC to transcription, leading to the rapid repair of DNA damage on the transcribed strand of transcribed genes66 The absence of orthologues of UvrABC in Arabidopsis renders the function of Mfd difficult to predict. The presence of Mfd but not UvrABC has been reported for only one other species, a bacterial endosymbiont of the pea aphid.

Other nuclear-encoded Arabidopsis DNA repair gene homologues are evolutionarily related to genes from α-Proteobacteria, and thus may be of mitochondrial descent. In particular, the six homologues of the alkyl-base glycosylase TagI appear to be the result of a large expansion in plants after transfer from the mitochondrial genome. Whether any of these TagI homologues function in the repair and maintenance of mitochondrial DNA has not been determined. More detailed phylogenetic analysis may reveal additional Arabidopsis RAR genes to be of organellar ancestry.

There are some notable absences of proteins important for RAR in other species, including alkyltransferases, MSH4, RPA3 and many components of TFIIH (TFB2, TFB3, TFB4, CCL1, Kin28). Nevertheless, Arabidopsis shows many similarities to the set of DNA repair genes found in other eukaryotes, and therefore offers an experimental system for determining the functions of many of these proteins, in part through characterization of mutants defective in DNA repair67.

Gene regulation

Eukaryotic gene expression involves many nuclear proteins that modulate chromatin structure, contribute to the basal transcription machinery, or mediate gene regulation in response to developmental, environmental or metabolic cues. As predicted by sequence similarity, more than 3,000 such proteins may be encoded by the Arabidopsis genome, suggesting that it has a comparable complexity of gene regulation to other eukaryotes. Arabidopsis has an additional level of gene regulation, however, with DNA methylation potentially mediating gene silencing and parental imprinting.

Plants have evolved several variations on chromatin remodelling proteins, such as the family of HD2 histone deacetylases68. Although Arabidopsis possesses the usual number of SNF2-type chromatin remodelling ATPases, which regulate the expression of nearly all genes, there are significant structural differences between yeast and metazoan SNF2-type genes and their orthologues in Arabidopsis. DDM1, a member of the SNF2 superfamily, and MOM1, a gene with similarity to the SNF2 family, are involved in transcriptional gene silencing in Arabidopsis. MOM1 has no clear orthologue in fungal or metazoan genomes.

Consistent with its methylated DNA, Arabidopsis possesses eight DNA methyltransferases (DMTs). Two of the three types are orthologous to mammalian DMT69 whereas one, chromomethyltransferase70, is unique to plants. No DMTs are found in yeast or C. elegans, although two DMT-like genes are found in Drosophila71. Arabidopsis also encodes eight proteins with methyl-DNA-binding domains (MBDs). Despite lacking methylated DNA, Drosophila encodes four MBD proteins and C. elegans has two. These differences in chromatin components are likely to reflect important differences in chromatin-based regulatory control of gene expression in eukaryotes (Supplementary Information Table 6; http://Ag.Arizona.Edu/chromatin/chromatin.html).

The Arabidopsis genome encodes transcription machinery for the three nuclear DNA-dependent RNA polymerase systems typical of eukaryotes ( Supplementary Information Table 6). Transcription by RNA polymerases II and III appears to involve the same machinery as is used in other eukarotes; however, most transcription factors for RNA polymerase I are not readily identified. Only two polymerase I regulators (other than polymerase subunits and TATA-binding protein) are apparent in Arabidopsis, namely homologues of yeast RRN3 and mouse TTF-1. All eukaryotes examined to date have distinct genes for the largest and second largest subunits of polymerase I, II and III. Unexpectedly, Arabidopsis has two genes encoding a fourth class of largest subunit and second-largest subunit (Supplementary Information Fig. 5). It will be interesting to determine whether the atypical subunits comprise a polymerase that has a plant-specific function. Four genes encoding single-subunit plastid or mitochondrial RNA polymerases have been identified in Arabidopsis (Supplementary Information Table 6). Genes for the bacterial β-, β′- and α-subunits of RNA polymerase are also present, as are homologues of various σ-factors, and these proteins may regulate chloroplast gene expression. Mutations in the Sde-1 gene, encoding RNA-dependent RNA polymerase (RdRp), lead to defective post-transcriptional gene silencing72. We also identified five more closely related RdRp genes.

Our analysis, using both similarity searches and domain matches, has identified 1,709 proteins with significant similarity to known classes of plant transcription factors classified by conserved DNA-binding domains. This analysis used a consistent conservative threshold that probably underestimates the size of families of diverse sequence. This class of protein is the least conserved among all classes of known proteins, showing only 8–23% similarity to transcription factors in other eukaryotes (Fig. 2b). This reduced similarity is due to the absence of certain classes of transcription factors in Arabidopsis and large numbers of plant-specific transcription factors. We did not detect any members of several widespread families of transcription factors, such as the REL (Rel-like DNA-binding domain) homology region proteins, nuclear steroid receptors and forkhead-winged helix and POU (Pit-1, Oct- and Unc-8b) domain families of developmental regulators. Conversely, of 29 classes of Arabidopsis transcription factors, 16 appear to be unique to plants (Supplementary Information Table 6). Several of these, such as the AP2/EREBP-RAV, NAC and ARF-AUX/IAA families, contain unique DNA-binding domains, whereas others contain plant-specific variants of more widespread domains, such as the DOF and WRKY zinc-finger families and the two-repeat MYB family.

Functional redundancy among members of large families of closely related transcription factors in Arabidopsis is a significant potential barrier to their characterization73. For example, in the SHATTERPROOF and SEPALLATA families of MADS box transcription factors, all genes must be defective to produce visible mutant phenotypes74,75. These functionally redundant genes are found on the segmental duplications described above. Our analyses, together with the significant sequence similarity found in large families of transcription factors such as the R2R3-repeat MYB and WRKY families, suggest that strategies involving overexpression will be important in determining the functions of members of transcription factor families.

Arabidopsis has two or over three times more transcription factors than identified in Drosophila29 or C.elegans1, respectively. The significantly greater extent of segmental chromosomal and local tandem duplications in the Arabidopsis genome generates larger gene families, including transcription factors. The partly overlapping functions defined for a few transcription factors are also likely to be much more widespread, implicating many sequence-related transcription factors in the same cellular processes. Finally, the expanded number of genes involved in metabolism, defence and environmental interaction in Arabidopsis (Fig. 2a ), which have few counterparts in Drosophila and C. elegans , all require additional numbers and classes of transcription factors to integrate gene function in response to a vast range of developmental and environmental cues.

Cellular organization

Plant cells differ from animal cells in many features such as plastids, vacuoles, Golgi organization, cytoskeletal arrays, plasmodesmata linking cytoplasms of neighbouring cells, and a rigid polysaccharide-rich extracellular matrix—the cell wall. Because the cell wall maintains the position of a cell relative to its neighbours, both changes in cell shape and organized cell divisions, involving cytoskeleton reorganization and membrane vesicle targeting, have major roles in plant development. Plant cytokinesis is also unique in that the partitioning membrane is formed de novo by vesicle fusion. We compared the Arabidopsis genome with those of C. elegans, Drosophila and yeast to glimpse the genetic basis of plant-cell-specific features.

The principal components of the plant cytoskeleton are microtubules (MTs) and actin filaments (AFs); intermediate filaments (IFs) have not been described in plants. Arabidopsis appears to lack genes for cytokeratin or vimentin, the main components of animal IFs, but has several variants of actin, α- and β-tubulin. The Arabidopsis genome also encodes homologues of chaperones that mediate the folding of tubulin and actin polypeptides in yeast and animal cells, such as the prefoldin and cytosolic chaperonin complexes and tubulin-folding cofactors. The dynamic stability of MTs and AFs is influenced by MT-associated proteins and actin-binding proteins, respectively, several of which are encoded by Arabidopsis genes. These include the MT-severing ATPase katanin, AF-crosslinking/bundling proteins, such as fimbrins and villins, and AF-disassembling proteins, such as profilin and actin-depolymerizing factor/cofilin. The Arabidopsis proteome appears to lack homologues of proteins that, in animal cells, link the actin cytoskeleton across the plasma membrane to the extracellular matrix, such as integrin, talin, spectrin, α-actinin, vitronectin or vinculin. This apparent lack of ‘anchorage’ proteins is consistent with the different composition of the cell wall and with a prominence of cortical MTs at the expense of cortical AFs in plant cells.

Plant-specific cytoskeletal arrays include interphase cortical MTs mediating cell shape, the preprophase band marking the cortical site of cell division, and the phragmoplast assisting in cytokinesis76. Although plant cells lack structural counterparts of the yeast spindle pole body and the animal centrosome, Arabidopsis has homologues of core components of the MT-nucleating γ-tubulin ring complex, such as γ-tubulin, Spc97/hGCP2 and Spc98/hGCP3. Arabidopsis has numerous motor molecules, both kinesins and dyneins with associated dynactin complex proteins, which are presumably involved in the dynamic organization of MTs and in transporting cargo along MT tracks. There are also myosin motors that may be involved in AF-supported organelle trafficking. Essential features of the eukaryotic cytoskeleton appear to be conserved in Arabidopsis.

The Arabidopsis genome encodes homologues of proteins involved in vesicle budding, including several ARFs and ARF-related small G-proteins, large but not small ARF GEFs (adenosine ribosylation factor on guanine nucleotide exchange factor), adapter proteins, and coat proteins of the COP and non-COP types. Arabidopsis also has homologues of proteins involved in vesicle docking and fusion, including SNAP receptors (SNAREs), N-ethylmaleimide-sensitive factor (NSF) and Cdc48-related ATPases, accessory proteins such as Sec1 and soluble NSF attachment protein (SNAP), and Rab-type GTPases. The large number of Arabidopsis SNAREs can be grouped by sequence similarity to yeast and animal counterparts involved in specific trafficking pathways, and some have been localized to the trans-Golgi and the pre-vacuolar pathway77. Arabidopsis also has a receptor for retention of proteins in the endoplasmic reticulum, a cargo receptor for transport to the vacuole and several phragmoplastins related to animal dynamin GTPases. Thus, plant cells appear to use the same basic machinery for vesicle trafficking as yeast and animal cells.

Animal cells possess many functionally diverse small G-proteins of the Ras superfamily involved in signal transduction, AF reorganization, vesicle fusion and other processes. Surprisingly, Arabidopsis appears to lack genes for G-proteins of the Ras, Rho, Rac and Cdc42 subfamilies but has many Rab-type G-proteins involved in vesicle fusion and several Rop-type G-proteins, one of which has a role in actin organization of the tip-growing pollen tube78. The significance of this divergent amplification of different subfamilies of small G-proteins in plants and animals remains to be determined.

Arabidopsis possesses cyclin-dependent kinases (CDKs), including a plant-specific Cdc2b kinase expressed in a cell-cycle-dependent manner, several cyclin subtypes, including a D-type cyclin that mediates cytokinin-stimulated cell-cycle progression79, a retinoblastoma-related protein and components of the ubiquitin-dependent proteolytic pathway of cyclin degradation. In yeast and animal cells, chromosome condensation is mediated by condensins, sister chromatids are held together by cohesins such as Scc1, and metaphase–anaphase transition is triggered by separin/Esp1 endopeptidase proteolysis of Scc1 on APC-mediated degradation of its inhibitor, securin/Psd1. Related proteins are encoded by the Arabidopsis genome. Thus, the basic machinery of cell-cycle progression, genome duplication and segregation appears to be conserved in plants. By contrast, entry into M phase, M-phase progression and cytokinesis seem to be modified in plant cells. Arabidopsis does not appear to have homologues of Cdc25 phosphatase, which activates Cdc2 kinase at the onset of mitosis, or of polo kinase, which regulates M-phase progression in yeast and animals. Conversely, plant-specific mitogen-actived protein (MAP) kinases appear to be involved in cytokinesis.

Cytokinesis partitions the cytoplasm of the dividing cell. Yeast and animal cells expand the membrane from the surface towards the centre in a cleavage process supported by septins and a contractile ring of actin and type II myosin. By contrast, plant cytokinesis starts in the centre of the division plane and progresses laterally. A transient membrane compartment, the cell plate, is formed de novo by fusion of Golgi-derived vesicles trafficking along the phragmoplast MTs80. Consistent with the unique mode of plant cytokinesis, Arabidopsis appears to lack genes for septins and type II myosin. Conversely, cell-plate formation requires a cytokinesis-specific syntaxin that has no close homologue in yeast and animals. Although syntaxin-mediated membrane fusion occurs in animal cytokinesis and cellularization, the vesicles are delivered to the base of the cleavage furrow. Thus, the plant-specific mechanism of cell division is linked to conserved eukaryotic cell-cycle machinery.

Two main conclusions are suggested by this comparative analysis. First, Arabidopsis and eukaryotic cells have common features related to intracellular activities, such as vesicle trafficking, cytoskeleton and cell cycle. Second, evolutionarily divergent features, such as organization of the cytoskeleton and cytokinesis, appear to relate to the plant cell wall.

Development

The regulation of development in Arabidopsis, as in animals, involves cell–cell communication, hierarchies of transcription factors, and the regulation of chromatin state; however, there is no reason to suppose that the complex multicellular states of plant and animal development have evolved by elaborating the same general processes during the 1.6 billion years since the last common unicellular ancestor of plants and animals81,82. Our genome analyses reflect the long, independent evolution of many processes contributing to development in the two kingdoms.

Plants and animals have converged on similar processes of pattern formation, but have used and expanded different transcription factor families as key causal regulators. For example, segmentation in insects and differentiation along the anterior–posterior and limb axes in mammals both involve the spatially specific activation of a series of homeobox gene family members. The pattern of activation is causal in the later differentiation of body and limb axis regions. In plants the pattern of floral whorls (sepals, petals, stamens, carpels) is also established by the spatially specific activation of members of a family of transcription factors, but in this instance the family is the MADS box family. Plants also have homeobox genes and animals have MADS box genes, implying that each lineage invented separately its mechanism of spatial pattern formation, while converging on actions and interactions of transcription factors as the mechanism. Other examples show even greater divergence of plant and animal developmental control. Examples are the AP2/EREBP and NAC families of transcription factors, which have important roles in flower and meristem development; both families are so far found only in plants (Supplementary Information Table 6).

A similar story can be told for cell–cell communication. Plants do not seem to have receptor tyrosine kinases, but the Arabidopsis genome has at least 340 genes for receptor Ser/Thr kinases, belonging to many different families, defined by their putative extracellular domains ( Supplementary Information Table 7). Several families have members with known functions in cell–cell communication, such as the CLV1 receptor involved in meristem cell signalling, the S-glycoprotein homologues involved in signalling from pollen to stigma in self-incompatible Brassica species, and the BRI1 receptor necessary for brassinosteroid signalling83. Animals also have receptor Ser/Thr kinases, such as the transforming growth factor-β (TGF-β) receptors, but these act through SMAD proteins that are absent from Arabidopsis. The leucine-rich repeat (LRR) family of Arabidopsis receptor kinases shares its extracellular domain with many animal and fungal proteins that do not have associated kinase domains, and there are at least 122 Arabidopsis genes that code for LRR proteins without a kinase domain. Other Arabidopsis receptor kinase families have extracellular domains that are unfamiliar in animals. Thus, evolution is modular, and the plant and animal lineages have expanded different families of receptor kinases for a similar set of developmental processes.

Several Arabidopsis genes of developmental importance appear to be derived from a cyanobacteria-like genome (Supplementary Information Table 2), with no close relationship to any animal or fungal protein. One salient example is the family of ethylene receptors; another gene family of apparent chloroplast origin is the phytochromes—light receptors involved in many developmental decisions (see below). Whereas the land plant phytochromes show clear homology to the cyanobacterial light receptors, which are typical prokaryotic histidine kinases, the plant phytochromes are histidine kinase paralogues with Ser/Thr specificity84. Similarly to the ethylene receptors, the proteins that act downstream of plant phytochrome signalling are not found in cyanobacteria, and thus it appears that a bacterial light receptor entered the plant genome through horizontal transfer, altered its enzymatic activity, and became linked to a eukaryotic signal transduction pathway. This infusion of genes from a cyanobacterial endosymbiont shows that plants have a richer heritage of ancestral genes than animals, and unique developmental processes that derive from horizontal gene transfer.

Signal transduction

Being generally sessile organisms, plants have to respond to local environmental conditions by changing their physiology or redirecting their growth. Signals from the environment include light and pathogen attack, temperature, water, nutrients, touch and gravity. In addition to local cellular responses, some stimuli are communicated across the plant body, with plant hormones and peptides acting as secondary messengers. Some hormones, such as auxin, are taken up into the cell, whereas others, such as ethylene and brassinosteroids, and the peptide CLV3, act as ligands for receptor kinases on the plasma membrane. No matter where the signal is perceived by the cell, it is transduced to the nucleus, resulting in altered patterns of gene expression.

Comparative genome analysis between Arabidopsis, C. elegans and Drosophila supports the idea that plants have evolved their own pathways of signal transduction85. None of the components of the widely adopted signalling pathways found in vertebrates, flies or worms, such as Wingless/Wnt, Hedgehog, Notch/lin12, JAK/STAT, TGF-β/SMADs, receptor tyrosine kinase/Ras or the nuclear steroid hormone receptors, is found in Arabidopsis. By contrast, brassinosteroids are ligands of the BRI1 Ser/Thr kinase, a member of the largest recognizable class of transmembrane sensors encoded by 340 receptor-like kinase (RLK) genes in the Arabidopsis genome (Supplementary Information Table 7). With a few notable exceptions, such as CLV1, the types of ligands sensed by RLKs are completely unknown, providing an enormous future challenge for plant biologists. G-protein-coupled receptors (GPCRs)/ seven-transmembrane proteins are an abundant class of proteins in mammalian genomes, instrumental in signal transduction. INTERPRO detected 27 GPCR-related domains in Arabidopsis ( Supplementary Information Table 1), although there is no direct experimental evidence for these. Arabidopsis contains a family of 18 seven-transmembrane proteins of the mildew resistance (MlO) class, several of which are involved in defence responses. Notably, only single Gα (GPA1) and Gβ (AGB1) subunits are found in Arabidopsis, both previously known86.

Although cyclic GMP has been proposed to be involved in signal transduction in Arabidopsis87, a protein containing a guanylate cyclase domain was not identified in our analyses. Nevertheless, cyclic nucleotide-binding domains were detected in various proteins, indicating that cNMPs may have a role in plant signal transduction. Thus, although cNMP-binding domains appear to have been conserved during evolution, cNMP synthesis in Arabidopsis may have evolved independently.

We were unable to identify a protein with significant similarity to known Gγ subunits, but recent biochemical studies suggest that a protein with this functional capacity is likely to be present in plant cells (H. Ma, personal communication). Therefore, there is potential for the formation of only a single heterotrimeric G-protein complex; however, its functional interaction with any of the potential GPCR-related proteins remains to be determined.

Modules of cellular signal pathways from bacteria and animals have been combined and new cascades have been innovated in plants. A pertinent example is the response to the gaseous plant hormone ethylene88. Ethylene is perceived and its signal transmitted by a family of receptors related to bacterial-type two-component histidine kinases (HKs). In bacteria, yeast and plants, these proteins sense many extracellular signals and function in a His-to-Asp phosphorelay network89. In turn, these proteins physically interact with the genetically downstream protein CTR1, a Raf/MAPKKK-related kinase, revealing the juxtaposition of bacterial-type two-component receptors and animal-type MAP kinase cascades. Unlike animals, however, Arabidopsis does not seem to have a Ras protein to activate the MAP kinase cascade. MAP kinases are found in abundance in Arabidopsis: we identified 20, a higher number than in any other eukaryote. As potentially counteracting components, we found 70 putative PP2C protein phosphatases. Although this group is largely uncharacterized functionally, several members are related to ABI1/ABI2, key negative regulators in the signalling pathway for the plant hormone abscisic acid. Additional components of the His-to-Asp phosphorelay system were also found in Arabidopsis, including authentic response regulators (ARRs), pseudoresponse regulators (PRRs) and phosphotransfer intermediate protein (HPt)90. We found 11 HKs in the proteome (3 new), 16 RRs (2 new) and 8 PRRs (2 new). The biological roles of most ARRs, PRRs and HPts are largly unknown, but several have been found to have diverse functions in plants, including transcriptional activation in response to the plant hormone cytokinin91, and as components of the circadian clock92.

Plants seem to have evolved unique signalling pathways by combining a conserved MAP kinase cascade module with new receptor types. In many cases, however, the ligands are unknown. Conversely, some known signalling molecules, such as auxin, are still in search of a receptor. Auxin signalling may represent yet another plant-specific mode of signalling, with protein degradation through the ubiquitin-proteasome pathway preceding altered gene expression. With many Arabidopsis genes encoding components of the ubiquitin-proteasome pathway, elimination of negative regulators may be a more widespread phenomenon in plant signalling.

Recognizing and responding to pathogens

Plants are constantly exposed to pests, parasites and pathogens and have evolved many defences. In mammals, polymorphism for parasite recognition encoded in the MHC genes contributes to resistance. In plants, disease resistance (R) genes that confer parasite recognition are also extremely polymorphic. This polymorphism has been proposed to restrict parasites, and its absence may explain the breakdown of resistance in crop monocultures93. In contrast to MHC genes, plant resistance genes are found at several loci, and the complete genome sequence enables analysis of their complement and structure. Parasite recognition by resistance genes triggers defence mechanisms through various signalling molecules, such as protein kinases and adapter proteins, ion fluxes, reactive oxygen intermediates and nitric oxide. These halt pathogen colonization through transcriptional activation of defence genes and a form of programmed cell death called the hypersensitive response94. The Arabidopsis genome contains diverse resistance genes distributed at many loci, along with components of signalling pathways, and many other genes whose role in disease resistance has been inferred from mutant phenotypes.

Most resistance genes encode intracellular proteins with a nucleotide-binding (NB) site typical of small G proteins, and carboxy-terminal LRRs95. Their amino termini either carry a TIR domain, or a putative coiled coil (CC). There are 85 TIR–NB–LRR resistance genes at 64 loci, and 36 CC–NB–LRR resistance genes at 30 loci. Some NB–LRR resistance genes express neither obvious TIR nor CC domains at their N termini. This potential class is present seven times, at six loci. There are 15 truncated TIR–NB genes that lack an LRR at 10 loci, often adjacent to full TIR–NB–LRR genes. There are also six CC–NB genes, at five loci. These truncated products may function in resistance. Intriguingly, two TIR–NB–LRR genes carry a WRKY domain, found in transcription factors that are implicated in plant defence, and one of these also encodes a protein kinase domain.

Resistance gene evolution may involve duplication and divergence of linked gene families36; however, most (46) resistance genes are singletons; 50 are in pairs, 21 are in 7 clusters of 3 family members, with single clusters of 4, 5, 7, 8 and 9 members, respectively. Of the non-singletons, 60% of pairs are in direct repeats, and 40% are in inverted repeats. Resistance genes are unevenly distributed between chromosomes, with 49 on chromosome 1; 2 on chromosome 2; 16 on chromosome 3; 28 on chromosome 4; and 55 on chromosome 5.

In other plant species, resistance genes encode both transmembrane receptors for secreted pathogen products and protein kinases, and some other classes are also found. The Cf genes in tomato encode extracellular LRRs with a transmembrane domain and short cytoplasmic domain. Mutation in an Arabidopsis homologue, CLAVATA2, results in enlarged meristems, but to date no resistance function has been assigned to the 30 Arabidopsis CLV2 homologues. CLAVATA1, a transmembrane LRR kinase, is also required for meristem function. Xa21, a rice LRR-kinase, confers Xanthomonas resistance, and the Arabidopsis FLS2 LRR kinase confers recognition of flagellin. It has been proposed that CLV1 and CLV2 function as a heterodimer; perhaps this is also true for Xa21, FLS2 and Cf proteins. There are 174 LRR transmembrane kinases in Arabidopsis, with only FLS2 assigned a role in resistance. A unique resistance gene, beet Hs1pro-1, which confers nematode resistance, has two Arabidopsis homologues.

The tomato Pto Ser/Thr kinase acts as a resistance protein in conjunction with an NB–LRR protein, so similar kinases might do the same for Arabidopsis NB–LRR proteins. There are 860 Ser/Thr kinases in the Arabidopsis sequence. Fifteen of these share 50% identity over the Pto-aligned region. The Toll pathway in Drosophila and mammals regulates innate immune responses through LRR/TIR domain receptors that recognize bacterial lipopolysaccharides96. Pto is highly homologous to Drosophila PELLE and mammalian IRAK protein kinases that mediate the TIR pathway.

Additional genes have been defined that are required for resistance by our analysis of the genome sequence. The ndr1 mutation defines a gene required by the CC–NB–LRR gene RPS2 and RPM1. NDR1 is 1 of 28 Arabidopsis genes that are similar both to each other and to the tobacco HIN1 gene that is transcriptionally induced early during the hypersensitive response. EDS1 is a gene required for TIR–NB–LRR function, and like PAD4, encodes a protein with a putative lipase motif. EDS1, PAD4 and a third gene comprise the EDS1/PAD4 family. The NPR1/NIM1/SAI1 gene is required for systemic acquired resistance, and we found five additional NPR1 homologues. Recessive mutations at both the barley Mlo and Arabidopsis LSD1 loci confer broad-spectrum resistance and derepress a cell-death program. There are at least 18 Mlo family members that resemble heterotrimeric GPCRs in Arabidopsis, and only two LSD1 homologues.

One of the earliest responses to pathogen recognition is the production of reactive oxygen intermediates. This involves a specialized respiratory burst oxidase protein that transfers an electron across the plasma membrane to make superoxide. Arabidopsis encodes eight apparently functional gp91 homologues, called Atrboh genes. Unlike gp91, they all carry an 300 amino-acid N-terminal extension carrying an EF-hand Ca2+-binding domain. In mammals, activation of the respiratory oxidative burst complex in the neutrophil, which includes gp91, requires the action of Rac proteins. As no Rac or Ras proteins are found in Arabidopsis , members of the large rop family of G proteins may carry this out. Similarly, we did not detect any Arabidopsis homologues of other mammalian respiratory burst oxidase components (p22, p47, p67, p40).

There are no clear homologues of many mammalian defence and cell-death control genes. Although nitric oxide production is involved in plant defence, there is no obvious homologue of nitric oxide synthase. Also absent are apparent homologues of the REL domain transcription factors involved in innate immunity in both Drosophila and mammals. We found no similarity to proteins involved in regulating apoptosis in animal cells, such as classical caspases, bcl2/ced9 and baculovirus p35. There are, however, 36 cysteine proteases. There are also eight homologues of a newly defined metacaspase family97, two of which, along with LSD1, have a clear GATA-type zinc-finger.

Photomorphogenesis and photosynthesis

Because nearly all plants are sessile and most depend on photosynthesis, they have evolved unique ways of responding to light. Light serves as an energy source, as well as a trigger and modulator of complex developmental pathways, including those regulated by the circadian clock. Light is especially important during seedling emergence, where it stimulates chlorophyll production, leaf development, cotyledon expansion, chloroplast biogenesis and the coordinated induction of many nuclear- and chloroplast-encoded genes, while at the same time inhibiting stem growth. The goal of this process, called photomorphogenesis, is the establishment of a body plan that allows the plant to be an efficient photosynthetic machine under varying light conditions98. The signal transduction cascade leading to light-induced responses begins with the activation of photoreceptors. Next, the light signal is transduced via positively and negatively acting nuclear and cytoplasmic proteins, causing activation or derepression of nuclear and chloroplast-encoded photosynthetic genes and enabling the plant to establish optimal photoautotrophic growth. Although genetic and biochemical studies have defined many of the components in this process, the genome sequence provides an opportunity to identify comprehensively Arabidopsis genes involved in photomorphogenesis and the establishment of photoautotrophic growth. We identified at least 100 candidate genes involved in light perception and signalling, and 139 nuclear-encoded genes that potentially function in photosynthesis.

The roles have been described of only 35 of the 100 candidate photomorphogenic genes (Supplementary Information Table 8). All of the light photoreceptors had been discovered previously, including five red/far-red absorbing phytochromes (PHYA-E), two blue/ultraviolet-A absorbing cryptochromes (CRY1 and CRY2), one blue-absorbing phototropin (NPH1) and one NPH1-like (or NPL1). In contrast, we uncovered many new proteins similar to the photomorphogenesis regulators COP/DET/FUS, PKS1, PIF3, NDPK2, SPA1, FAR1, GIGANTEA, FIN219, HY5, CCA1, ATHB-2, ZEITLUPE, FKF1, LKP1, NPH3 and RPT2.

Both the phytochromes and NPH1 contain chromophores for light sensing coupled to kinase domains for signal transmission. Phytochromes have an N-terminal chromophore-binding domain, two PAS domains, and a C-terminal Ser/Thr kinase domain99, whereas NPH1 has two LOV domains (members of the PAS domain superfamily) for flavin mononucleotide binding and a C-terminal Ser/Thr kinase domain100. PAS domains potentially sense changes in light, redox potential and oxygen energy levels, as well as mediating protein–protein interactions99,100. We searched for uncharacterized proteins with the combination of a kinase domain and either a phytochrome chromophore-binding site or PAS domains. Although we found no new phytochrome-like genes, we did identify four predicted proteins that contain PAS and kinase domains (Supplementary Information Fig. 6). These proteins share 80% amino-acid identity, but, unlike NPH1 and NPL1, have only one PAS domain. The combination of potential signal sensing and transmitting domains makes it tempting to speculate that these proteins may be receptors for light or other signals.

Our screen included searches for components of photosynthetic reaction centres and light-harvesting complexes, enzymes involved in CO2 fixation and enzymes in pigment biosynthesis. We identified 11 core proteins of photosystem I, including the eukaryotic-specific components PsaG and PsaH101, and 8 photosystem II proteins, including a single member (psbW) of the photosystem II core. We also found 26 proteins similar to the Chlorophyll-a/b binding proteins (8 Lhca and 18 Lhcb). Of the seven subunits of the cytochrome b6f complex (PetA–D, PetG, PetL, PetM), only one ( PetC) was found in the nuclear genome, whereas the remainder are probably encoded in the chloroplast. Similarly, of the nine subunits of the chloroplast ATP synthase complex, three are encoded in the nucleus, including the II- , γ- and δ-subunits; the remaining subunits (I, III, IV, α, β, ε) are encoded in the chloroplast102. Ten genes were related to the soluble components of the electron transfer chain, including two plastocyanins, five ferredoxins and three ferredoxin/NADP oxidoreductases. Forty genes are predicted to have a role in CO2 fixation, including all of the enzymes in the Calvin–Benson cycle. For pigment biosynthesis, 16 genes in chlorophyll biosynthesis and 31 genes in carotenoid biosynthesis were found (Supplementary Information Table 8). Our analyses have identified several potential components of the light perception pathway, and have revealed the complex distribution of components of the photosynthetic apparatus between nuclear and plastid genomes.

Metabolism

Arabidopsis is an autotrophic organism that needs only minerals, light, water and air to grow. Consequently, a large proportion of the genome encodes enzymes that support metabolic processes, such as photosynthesis, respiration, intermediary metabolism, mineral acquisition, and the synthesis of lipids, fatty acids, amino acids, nucleotides and cofactors103. With respect to these processes, Arabidopsis appears to contain a complement of genes similar to those in the photoautotropic cyanobacterium Synechocystis 45, but, whereas Synechocystis generally has a single gene encoding an enzyme, Arabidopsis frequently has many. For example, Arabidopsis has at least seven genes for the glycolytic enzyme pyruvate kinase, with an additional five for pyruvate kinase-like proteins. Whatever the reason for this high level of redundancy, it varies from gene to gene in the same pathway; the 11 enzymes of glycolysis are encoded by up to 51 genes that are present in as few as one or as many as eight copies. Similarly, of the 59 genes encoding proteins involved in glycerolipid metabolism, 39 are represented by more than one gene104. Genome duplication and expansion of gene families by tandem duplication have contributed to this diversity.

This high degree of apparent structural redundancy does not necessarily imply functional redundancy. For instance, although there are seven genes for serine hydroxymethyltransferase, a mutation in the gene for the mitochondrial form completely blocks the photorespiratory pathway105. Although there are 12 genes for cellulose synthase, mutations in at least 2 of the 12 confer distinct phenotypes because of tissue-specific gene expression106.

The metabolome of Arabidopsis differs from that of cyanobacteria, or of any other organism sequenced to date, by the presence of many genes encoding enzymes for pathways that are unique to vascular plants. In particular, although relatively little is known about the enzymology of cell-wall metabolism, more than 420 genes could be assigned probable roles in pathways responsible for the synthesis and modification of cell-wall polymers. Twelve genes encode cellulose synthase, and 29 other genes encode 6 families of structurally related enzymes thought to synthesize other major polysaccharides106. Roughly 52 genes encode polygalacturonases, 20 encode pectate lyases and 79 encode pectin esterases, indicating a massive investment in modifying pectin. Similarly, the presence of 39 β-1,3-glucanases, 20 endoxyloglucan transglycosylases, 50 cellulases and other hydrolases, and 23 expansins reflects the importance of wall remodelling during growth of plant cells. Excluding ascorbate and glutathione peroxidases, there are 69 genes with significant similarity to known peroxidases and 15 laccases (diphenol oxidases). Their presence in such abundance indicates the importance of oxidative processes in the synthesis of lignin, suberin and other cell-wall polymers. The high degree of apparent redundancy in the genes for cell-wall metabolism might reflect differences in substrate specificity by some of the enzymes.

The high degree of apparent redundancy in the genes for cell wall metabolism might reflect differences in substrate specificity by some of the enzymes. It is already known that cell types have different wall compositions, which may require that the relevant enzymes be subject to cell-type-specific transcriptional regulation. Of the 40 or so cell types that plants make, almost all can be identified by unique features of their cell wall107. A large number of genes involved in wall metabolism have yet to be defined. Although more than 60 genes for glycosyltransferases can be found in the genome sequence, most of these are probably involved in protein glycosylation or metabolite catabolism and do not seem to be adequate to account for the polysaccharide complexity of the wall. For instance, at least 21 enzymes are required just to produce the linkages of the pectic polysaccharide RGII, and none of these enzymes has been identified at present. Thus, if these and related enzymes involved in the synthesis of other cell-wall polymers are also represented by multiple genes, a substantial number of the genes of currently unknown function may be involved in cell-wall metabolism.

Higher plants collectively synthesize more than 100,000 secondary metabolites. Because flowering plants are thought to have similar numbers of genes, it is apparent that a great deal of enzyme creation took place during the evolution of higher plants. An important factor in the rapid evolution of metabolic complexity is the large family of cytochrome P450s that are evident in Arabidopsis (Supplementary Information Table 1). These enzymes represent a superfamily of haem-containing proteins, most of which catalyse NADPH- and O2-dependent hydroxylation reactions. Plant P450s participate in myriad biochemical pathways including those devoted to the synthesis of plant products, such as phenylpropanoids, alkaloids, terpenoids, lipids, cyanogenic glycosides and glucosinolates, and plant growth regulators, such as gibberellins, jasmonic acid and brassinosteroids. Whereas Arabidopsis has 286 P450 genes, Drosophila has 94, C. elegans has 73 and yeast has only 3. This low number in yeast indicates that there are few reactions of basic metabolism that are catalysed by P450s. It seems likely that many animal P450s are involved in detoxification of compounds from food plant sources. The role of endogenous enzymes is poorly understood; only a few dozen P450 enzymes from plants have been characterized to any extent. The discrepancy between the number of known P450-catalysed reactions and the number of genes suggests that Arabidopsis produces a relatively large number of metabolites that have yet to be identified.

In addition to the large number of cytochrome P450s, Arabidopsis has many other genes that suggest the existence of pathways or processes that are not currently known. For instance, the presence of 19 genes with similarity to anthranilate N-hydroxycinnamoyl/benzoyl transferase is currently inexplicable. This enzyme is involved in the synthesis of dianthramide phytoalexins in Caryophyllaceae and Gramineae. No phytoalexins of this class have been described in Arabidopsis as yet. Similarly, the presence of 12 genes with sequence similarity to the berberine bridge enzyme, ((S)-reticuline:oxygen oxidoreductase (methylene-bridge-forming); EC 1.5.3.9), and 13 genes with similarity to tropinone reductase, suggests that Arabidopsis may have the ability to produce alkaloids. In other plants, the berberine bridge enzyme transforms reticuline into scoulerine, a biosynthetic precursor to a multitude of species-specific protopine, protoberberine and benzophenanthridine alkaloids. The discovery of these and many other intriguing genes in the Arabidopsis genome has created a wealth of new opportunities to understand the metabolic and structural diversity of higher plants.

Concluding remarks

The twentieth century began with the rediscovery of Mendel's rules of inheritance in pea108, and it ends with the elucidation of the complete genetic complement of a model plant, Arabidopsis. The analysis of the completed sequence of a flowering plant reported here provides insights into the genetic basis of the similarities and differences of diverse multicellular organisms. It also creates the potential for direct and efficient access to a much deeper understanding of plant development and environmental responses, and permits the structure and dynamics of plant genomes to be assessed and understood.

Arabidopsis, C. elegans and Drosophila have a similar range of 11,000–15,000 different types of proteins, suggesting this is the minimal complexity required by extremely diverse multicellular eukaryotes to execute development and respond to their environment. We account for the larger number of gene copies in Arabidopsis compared with these other sequenced eukaryotes with two possible explanations. First, independent amplification of individual genes has generated tandem and dispersed gene families to a greater extent in Arabidopsis, and unequal crossing over may be the predominant mechanism involved. Second, ancestral duplication of the entire genome and subsequent rearrangements have resulted in segmental duplications. The pattern of these duplications suggests an ancient polyploidy event, and mutant analysis indicates that at least some of the many duplicate genes are functionally redundant. Their occurrence in a functionally diploid genetic model came as a surprise, and is reminiscent of the situation in maize, an ancient segmental allotetraploid. The remarkable degree of genome plasticity revealed in the large-scale duplications may be needed to provide new functions, as alternative promoters and alternative splicing appear to be less widely used in plants than they are in animals. Apart from duplicated segments, the overall chromosome structure of Arabidopsis closely resembles that of Drosophila; transposons and other repetitive sequences are concentrated in the heterochromatic regions surrounding the centromere, whereas the euchromatic arms are largely devoid of repetitive sequences. Conversely, most protein-coding genes reside in the euchromatin, although a number of expressed genes have been identified in centromeric regions. Finally, Arabidopsis is the first methylated eukaryotic genome to be sequenced, and will be invaluable in the study of epigenetic inheritance and gene regulation.

Unlike most animals, plants generally do not move, they can perpetuate indefinitely, they reproduce through an extended haploid phase, and they synthesize all their metabolites. Our comparison of Arabidopsis, bacterial, fungal and animal genomes starts to define the genetic basis for these differences between plants and other life forms. Basic intracellular processes, such as translation or vesicle trafficking, appear to be conserved across kingdoms, reflecting a common eukaryotic heritage. More elaborate intercellular processes, including physiology and development, use different sets of components. For example, membrane channels, transporters and signalling components are very different in plants and animals, and the large number of transcription factors unique to plants contrasts with the conservation of many chromatin proteins across the three eukaryotic kingdoms. Unexpected differences between seemingly similar processes include the absence of intracellular regulators of cell division (Cdc25) and apoptosis (Bcl-2). On the other hand, DNA repair appears more highly conserved between plants and mammals than within the animal kingdom, perhaps reflecting common factors such as DNA methylation. Our analysis also shows that many genes of the endosymbiotic ancestor of the plastid have been transferred to the nucleus, and the products of this rich prokaryotic heritage contribute to diverse functions such as photoautotrophic growth and signalling.

The sequence reported here changes the fundamental nature of plant genetic analysis. Forward genetics is greatly simplified as mutations are more conveniently isolated molecularly, but at the same time extensive gene duplications mean that functional redundancy must be taken into account. At a biochemical level, the specificity conferred by nucleotide sequence, and the completeness of the survey allow complex mixtures of RNA and protein to be resolved into their individual components using micro-arrays and mass spectrometry. This specificity can also be used in the parallel analysis of genome-wide polymorphisms and quantitative traits in natural populations109. Looking ahead, the challenge of determining the function of the large set of predicted genes, many of which are plant-specific, is now a clear priority, and multinational programs have been initiated to accomplish this goal using site-selected mutagenesis among the the necessary tools110. Finally, productive paths of crop improvement, based on enhanced knowledge of Arabidopsis gene function, will help meet the challenge of sustaining our food supply in the coming years.

Note added in proof: at the time of publication 17 centromeric BACs and 5 sequence gaps in chromosome arms are being sequenced.

Methods

The three centres used similar annotation approaches involving in silico gene-finding methods, comparison to EST and protein databases, and manual reconciliation of that data. Gene finding involved three steps: (1) analysis of BAC sequences using a computational gene finder; (2) alignment of the sequence to the protein and EST databases; (3) assignment of functions to each of the genes. Genscan111, GeneMark.HMM112, Xgrail113 Genefinder (P. Green, unpublished software) and GlimmerA114 were used to analyse BAC sequences. All of these systems were specially trained for Arabidopsis genes. Splice sites were predicted using NetGene2115, Splice Predictor116 and GeneSplicer (M. Pertea and S. Salzberg, unpublished software). For the second step, BACs were aligned to ESTs and to the Arabidopsis gene index117 using programs such as DDS/GAP2118 or BLASTN119. Segmental duplications were analysed and displayed using a modified version of DIALIGN2 (ref. 120).