The Arabidopsis Information Resource (TAIR): gene structure and function annotation

Arabidopsis annotation past and present

Although the Arabidopsis genome sequence was completed in 2000, much work remains to be done to incorporate all available experimental data on gene structure and function into the genome annotation. The Arabidopsis genome is the only dicot genome finished and annotated to a high standard, and (with rice) one of only two finished, rather than draft quality, plant genome sequences. Because annotation of new genomes is in large measure based on the annotation of existing complete genomes, improvements to the Arabidopsis genome annotation will directly aid the annotation of future plant genomes.

TAIR assumed primary responsibility for updating the Arabidopsis annotation following the Institute for Genomic Research's (TIGR's) final genome release in 2004 (5). Since the original Arabidopsis genome annotation in which 25 498 genes were reported (6), the number of annotated genes has steadily increased as new sequencing and array-based technologies have provided evidence for many previously unannotated genes (Table 1). In particular, new sequence data deposited in the nucleotide sequence databases (EMBL/GenBank/DDBJ), along with user submissions from the Arabidopsis community, has resulted in the addition of over a thousand new genes since the final TIGR release in January 2004.

TAIR annotation pipeline

The TAIR gene structure annotation pipeline incorporates both manual and automated updates. Automated updates are carried out using the Program to Assemble Spliced Alignments (PASA) optimized for Arabidopsis (7). Updates are manually reviewed before incorporation by examining available transcript evidence. For the latest genome release (TAIR7), 581 (25%) of 2298 suggested updates (excluding simple UTR extensions) were rejected. Rejected updates included cases where genomic sequencing errors or misalignments of poor-quality ESTs or other aberrant transcripts resulted in incorrect gene structures. Additional rejected updates included gene splits based on partial cDNAs. The occurrence of such a large fraction of incorrect automated updates highlights the continued benefits of manual annotation to achieve a gold standard of annotation.

In addition to using the PASA pipeline to fold in new transcript evidence, TAIR has employed a targeted approach to identify and resolve structural errors in existing gene models. For example, some UTRs were falsely extended in previous releases based on ESTs of ambiguous orientation (i.e. non-splicing transcripts without poly(A) features, which originated from genes on the opposite strand). We manually reviewed candidate genes potentially containing falsely extended UTRs and corrected a total of 1098 gene models (909 genes) for the TAIR7 release. This has resulted in several hundred array elements no longer mapping to previously associated genes. In addition, the average 3' UTR length decreased from 250 to 233 bp in the TAIR7 release, contrary to the usual expectation that annotated UTRs are extended between releases due to the availability of more transcript data. In contrast, the average 5′ UTR size increased from 139 to 146 bp for this release.

TAIR7 genome release

The latest genome release, TAIR7, contains annotations for 27 029 protein-coding genes (including mitochondrial and chloroplast genomes), 3889 pseudogenes or transposable elements and 1123 ncRNAs (32 041 genes in all, with 37 019 gene models 31 921 of which are protein-coding models). Of the 27 029 protein-coding genes, 3799 (14%) are annotated with alternatively spliced gene models and 70% of all protein-coding models have annotated 5′ and 3′ UTRs. Seventy-one percent (22 596) of model structures are confirmed by transcript data (where every coding exon is supported by an A. thaliana EST or cDNA). A further 5819 gene models are partially supported. Thus, a total of 28 415 (89%) protein-coding gene models have at least partial transcript support, an increase of 5001 compared to November 2005 (the time of the TAIR6 release). In the main, this is due to the increased number of EST sequences deposited in GenBank, particularly those originating from high-throughput sequencing technologies (8). While the total number of genes continues to rise with each release (Table 1) (681 new genes were added for the latest release), the most common updates are refinements to existing gene structures. For the TAIR7 release 10 792 gene structures were updated, of which 797 gene models had updated coding exons. A total of 14 050 exons were modified and 828 new exons were incorporated. There were 41 gene splits and 34 gene merges. Overall, one-third of all existing TAIR6 genes (10 098 genes) were updated for the TAIR7 release.

Expanding the Arabidopsis annotation by inclusion of small proteins and non-coding transcripts

Whereas the original Arabidopsis genome annotation focused solely on identifying protein-coding genes, subsequent re-annotations have added additional gene classes (i.e. pseudogenes, ncRNAs and transposable elements). However, certain gene classes may still be underrepresented. Transcriptome sequencing and whole-genome tiling array studies have revealed significant levels of expression in unannotated ‘intergenic’ regions (9,10). Not all such transcripts will be of functional significance but at least some are likely to represent unannotated protein-coding genes, ncRNAs or transposable elements. The problem of false-negative prediction is particularly serious for smaller CDSs where the decreasing signal-to-noise ratio makes distinguishing real genes from biologically meaningless open reading frames (ORFs) a more problematic issue (11). In order to reduce the number of short false-positive gene predictions TIGR applied a minimum cut-off of 110 amino acids (5). Unannotated protein-coding genes are therefore likely to fall below this threshold. Some previously unannotated small genes including 467 cysteine-rich peptides (12,13) have now been incorporated into the most recent genome releases (TAIR6 and TAIR7) while others are currently being evaluated for potential inclusion in the forthcoming release (14). In addition many unannotated transcripts excluded from earlier releases due to a lack of clear coding potential or assignment within one of the small RNA classes have now been incorporated. A total 213 genes were added as gene type ‘other_RNA’ where the gene cannot be assigned an unambiguous ORF and may therefore function as an ncRNA. One hundred and eighteen of these gene models overlap antisense to existing protein-coding genes and may represent natural antisense genes.

Alternatively spliced genes

While our latest annotation includes 1002 new splice variants, numerous sequences in GenBank that represent alternatively spliced transcripts with disrupted ORFs are not presently incorporated into TAIR gene structures. Many of these transcripts contain retained introns or alternative splice sites that generate premature termination codons (PTCs) and heavily truncated peptides. It remains an open question if such transcripts encode functional peptides. Some will undoubtedly represent mistakes by the splicing machinery or cloning and sequencing errors, while others may have regulatory roles via the nonsense-mediated decay (NMD) pathway, a surveillance mechanism that selectively degrades nonsense mRNAs. Coupling of regulated alternative splicing and NMD may therefore provide an alternative mechanism for regulating protein expression (15). We plan to incorporate and categorize these and other unusual transcripts encoding fused proteins or dicistronic transcripts containing two complete ORFs in future releases.

Functional annotation

At the level of gene function, we continue to refine our annotations using information from the literature and community-submitted annotations, as well as sequence similarity and other computational methods. Of the 28 152 genes in TAIR (excluding 3889 transposon genes and pseudogenes and 793 genetically defined, unsequenced genes), ∼60% have been annotated to a GO molecular function, 50% to a GO biological process and 49% to a GO cellular component (excluding annotations to the ‘unknown’ terms). In addition to updating GO annotations, TAIR updates the gene description field either manually, using literature data, or computationally, using BLAST and InterProScan to identify similar proteins and protein domains. Approximately 4000 genes currently have similarity only to uncharacterized proteins (i.e. hypothetical, predicted, unknown, etc), while 758 have no significant protein similarity to any proteins deposited in GenBank. Of these, 286 in addition have no supporting EST/cDNA evidence and may represent erroneous gene predictions incorporated in earlier releases.

Future releases

Future TAIR genome releases will contain corrections to the chromosome sequences based on newly available sequence data, improved annotation of endogenous transposable elements and pseudogenes, more complete annotation of splice variants including those lacking coding potential and utilization of genome comparisons to further refine the existing gene structures, particularly those with little or no transcript support. We plan to produce new genome releases at least once per year, with the next release (TAIR8) expected in early 2008.

• 1.Stein L, Mungall C, Shu S, Caudy M, Mangone M, Day A, Nickerson E, Stajich J, Harris T, et al. The generic genome browser: a building block for a model organism database. Genome Res. 2002;12:1599–1610. doi: 10.1101/gr.403602. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 2.Frazer KA, Pachter L, Poliakov A, Rubin EM, Dubchak I. VISTA: computational tools for comparative genomics. Nucleic Acids Res. 2004;32:W273–W279. doi: 10.1093/nar/gkh458. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 3.Wilkinson MD, Links M. BioMOBY: an open source biological web services proposal. Brief Bioinform. 2002;3:331–341. doi: 10.1093/bib/3.4.331. [DOI] [PubMed] [Google Scholar]
• 4.Hull D, Wolstencroft K, Stevens R, Goble C, Pocock M, Li P, Oinn T. Taverna: a tool for building and running workflows of services. Nucleic Acids Res. 2006;34:729–732. doi: 10.1093/nar/gkl320. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 5.Haas BJ, Wortman JR, Ronning CM, Hannick LI, Smith RK, Jr, Maiti R, Chan AP, Yu C, Farzad M, et al. Complete reannotation of the Arabidopsis genome: methods, tools, protocols and the final release. BMC Biol. 2005;3:7. doi: 10.1186/1741-7007-3-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 6.Arabidopsis Genome Initiative. Analysis of the genome sequence of the flowering plant Arabidopsis thaliana. Nature. 2000;408:796–815. doi: 10.1038/35048692. [DOI] [PubMed] [Google Scholar]
• 7.Haas BJ, Delcher AL, Mount SM, Wortman JR, Smith RK, Jr, Hannick LI, Maiti R, Ronning CM, Rusch DB, et al. Improving the Arabidopsis genome annotation using maximal transcript alignment assemblies. Nucleic Acids Res. 2003;31:5654–5666. doi: 10.1093/nar/gkg770. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 8.Weber AP, Weber KL, Carr K, Wilkerson C, Ohlrogge JB. Sampling the Arabidopsis transcriptome with massively parallel pyrosequencing. Plant Physiol. 2007;144:32–42. doi: 10.1104/pp.107.096677. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 9.Yamada K, Lim J, Dale JM, Chen H, Shinn P, Palm CJ, Southwick AM, Wu HC, Kim C, et al. Empirical analysis of transcriptional activity in the Arabidopsis genome. Science. 2003;302:842–846. doi: 10.1126/science.1088305. [DOI] [PubMed] [Google Scholar]
• 10.Stolc V, Samanta MP, Tongprasit W, Sethi H, Liang S, Nelson DC, Hegeman A, Nelson C, Rancour D, et al. Identification of transcribed sequences in Arabidopsis thaliana by using high-resolution genome tiling arrays. Proc. Natl. Acad. Sci. USA. 2005;102:4453–4458. doi: 10.1073/pnas.0408203102. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 11.Wang J, Li S, Zhang Y, Zheng H, Xu Z, Ye J, Yu J, Wong GK. Vertebrate gene predictions and the problem of large genes. Nat. Rev. Genet. 2003;4:741–749. doi: 10.1038/nrg1160. [DOI] [PubMed] [Google Scholar]
• 12.Silverstein KA, Graham MA, Paape TD, VandenBosch KA. Genome organization of more than 300 defensin-like genes in Arabidopsis. Plant Physiol. 2005;138:600–610. doi: 10.1104/pp.105.060079. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 13.Silverstein KA, Moskal WA, Jr, Wu HC, Underwood BA, Graham MA, Town CD, VandenBosch KA. Small cysteine-rich peptides resembling antimicrobial peptides have been under-predicted in plants. Plant J. 2007;51:262–280. doi: 10.1111/j.1365-313X.2007.03136.x. [DOI] [PubMed] [Google Scholar]
• 14.Hanada K, Zhang X, Borevitz JO, Li WH, Shiu SH. A large number of novel coding small open reading frames in the intergenic regions of the Arabidopsis thaliana genome are transcribed and/or under purifying selection. Genome Res. 2007;17 doi: 10.1101/gr.5836207. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 15.Lewis BP, Green RE, Brenner SE. Evidence for the widespread coupling of alternative splicing and nonsense-mediated mRNA decay in humans. Proc. Natl Acad. Sci. USA. 2003;100:189–192. doi: 10.1073/pnas.0136770100. [DOI] [PMC free article] [PubMed] [Google Scholar]