openSputnik--a database to ESTablish comparative plant genomics using unsaturated sequence collections.

The public expressed sequence tag collections are continually being enriched with high-quality sequences that represent an ever-expanding range of taxonomically diverse plant species. While these sequence collections provide biased insight into the populations of expressed genes available within individual species and their associated tissues, the information is conceivably of wider relevance in a comparative context. When we consider the available expressed sequence tag (EST) collections of summer 2004, most of the major plant taxonomic clades are at least superficially represented. Investigation of the five million available plant ESTs provides a wealth of information that has applications in modelling the routes of plant genome evolution and the identification of lineage-specific genes and gene families. Over four million ESTs from over 50 distinct plant species have been collated within an EST analysis pipeline called openSputnik. The ESTs were resolved down into approximately one million unigene sequences. These have been annotated using orthology-based annotation transfer from reference plant genomes and using a variety of contemporary bioinformatics methods to assign peptide, structural and functional attributes. The openSputnik database is available at http://sputnik.btk.fi.

INTRODUCTION

Complete genome sequencing has become the standard modus operandi for bacterial genomics, and tens of eukaryotic genomes have also been completely sequenced (see http://www.genomesonline.org). Plant genomics is, however, frequently hindered by the typically large and repetitive nature of the genome. Certain plant species have genome sizes that dwarf the human genome; the 1C genome size for broad bean (Vicia faba) is at least 26 000 Mb (Plant DNA C-values database), or over eight times the size of the human genome. The selection of candidate plant genomes for complete sequencing is, therefore, based on the scientific and anthropocentric value of the plant and the feasibility of a meaningful sequencing and assembly strategy. While several diverse plant species [Arabidopsis thaliana (1), Oryza sativa (2,3) and Populus trichocarpa] have been or will shortly be completely sequenced, the majority of plant genomes remain largely inaccessible. Arabidopsis and rice are certainly model plant systems but, are neither truly representative of any other given species nor are they general indicators for gene content across the whole plant kingdom. The first forays into comparative plant genomics using Arabidopsis and rice as reference genomes have demonstrated that there is a remarkable degree of underlying sequence diversity between these species (2,3). This firmly advocates the need to at least sample the protein-coding component of more taxonomically ‘exotic’ plant genomes.

cDNA preparation and expressed sequence tag (EST) sequencing remain a dominant methodology for accessing the protein coding (and expressed) portion of the genome. Many laboratories are independently sequencing very large numbers of sequences from a broad and bio-diverse spectrum of plant species (Figure 1). EST sequences retain their exalted status for several reasons [for a review see (4)].

Figure 1

A depiction of the phylogenetic relationships among the major plant lineages as published previously (23). The evolutionary tree has been overlaid with the names of plant species having large EST collections (>5000 sequences) that are available in the current release of openSputnik. The symbol ‘**’ denotes the plant groups where either small EST collections (>1000 ESTs) are available or as-yet unreleased sequences are known to exist. This figure reveals the taxonomic distribution of large plant EST collections, but also highlights the strong bias towards the agriculturally important species.

They are technically simple to produce and cheap to sequence.

ESTs provide a robust approximation of the expressed gene content of the parental genome under given sampling conditions and can be used for primitive expression profiling between tissues (5).

The extensive redundancy typical of EST collections also allows for the selection of putative molecular markers (6,7).

cDNAs may be used as a substrate for arraying, to create cDNA microarrays; this allows for true gene expression profiling (8).

With an excess of 5.4 million sequences from over 320 species, the current public plant EST sequence databases (EMBL release 80) (9) are a valuable and contextually rich but under-utilized resource. If we consider just the large EST collections with over 5000 ESTs, 5.1 million ESTs from 74 species are represented. These species, while highly biased towards the key plant taxonomic clades of the rosids, asterids and monocots, still contain representative species, from other key taxonomic groups. The species represented contain representatives of single cellularity—the red and brown algae and lower plants—gymnosperms, basal angiosperms and the angiosperms. With such a wealth of signals for investigation of the underlying genomic changes in gene-content, protein structures and domain composition, the EST collections surely deserve detailed analysis and investigation.

The openSputnik database has been designed as an interim platform for the exhaustive annotation and analysis of EST sequences in a comparative context. In addition to clustering sequences, a peptide sequence is identified, thus, providing a more sensitive target for the identification of functional and structural features. Sequences are placed in context with the currently available complete plant genomes and are associated with other clustered EST collections. The openSputnik database, thus, creates a platform upon which the intricate patterns of generalist house-keeping genes and lineage-specific gene families may be teased apart. The completed EST project annotations are available as a searchable web resource. While the provision of an integrated resource containing a diverse mixture of clustered and contextually placed unigene sequences is not unique [e.g. TIGR Gene Indices database (10), NCBI Unigenes (11) or PlantGDB at Iowa State University (12)], the openSputnik database is currently distinct in its focus towards functionally describing unigene sequences on the basis of both orthologous gene annotations and the application of bioinformatics methods for ab initio annotations.

IMPLEMENTATION AND STARTING MATERIAL

The openSputnik database has been programmed using the Java programming language and utilizes the PostgreSQL relational database management system to archive and retrieve sequences and their annotations. Therefore, openSputnik is largely platform-independent and has been implemented using a server–client model to allow for calculation in a distributed and heterogeneous computational environment. The methods implemented within openSputnik are described as functional objects and the analytical pathway is described as a directed acyclic graph (Figure 2). The current version of openSputnik utilizes the complete public plant EST collection that was available from the European Molecular Biology Laboratory (EMBL) at the start of Spring 2004 (EMBL release 78). A rule was imposed so that EST collections of at least 4500 sequences would be included. Over four million EST sequences representing 55 distinct plant species were identified using this rule. These sequences were loaded onto the openSputnik database schema.

Figure 2

A simplification of the directed acyclic graph that describes the analytical pipeline used to build the openSputnik database. As starting material, species-specific EMBL flat files are imported and all annotations are retained. This creates a sequence source ‘EST collection’. This source is used to derive two other annotative sources, the ‘UNIGENE collection’ and the ‘PEPTIDE collection’ (sources shown in red). When the sources have been built, they are annotated using a variety of methods highlighted in green. The analyses anchored to the schema are used to create derived annotations including Funcat and GO terms (shown in orange). All analyses are made available to the database user via the openZputnik interface.

SEQUENCE CLUSTERING

Prior to sequence clustering, ESTs were aggressively trimmed of any likely residual vector or polylinker sequences using the Crossmatch application (P. Green, unpublished data) and the National Center for Bioinformatics Information (NCBI) UniVec database. Sequences <55 nt in length were excluded at this stage. To prevent the aggregation of sequences on the basis of low complexity sequence islands, all low complexity sequences were masked using the RepeatBeater algorithm (Biomax informatics, Martinsried, Germany). The masked sequences were clustered into pools of related sequences using a suffix tree based approach (HPT2 algorithm; Biomax informatics). To encourage the aggregation of sequences, HPT2 was run using a similarity threshold of 0.7 and a number of network iterations equalling the number of masked ESTs. The resulting clusters were assembled into unigene sequences using the CAP3 algorithm with standard settings. Within the larger EST collections, some HPT2 identified clusters contain many members. To simplify the analysis, larger clusters were truncated to an arbitrary threshold of a maximum of 2500 ESTs. Some individual ESTs representing the most highly expressed genes were absent from their cognate unigenes.

PEPTIDE PREDICTION

It is probable that each derived unigene sequence represents an expressed and properly spliced mRNA. Extensive amounts of either 5′-untranslated region (5′-UTR) or 3′-UTR may exist within the unigene sequences. The identification of a meaningful peptide sequence lends value to the dataset by allowing us to exclude sequences of low protein-coding potential, and additionally allows the use of peptide-annotation algorithms. ESTScan (13) models have been trained for each of the underlying species. Training data were produced by identifying probable open reading frame (ORF) sequences from a BLASTX (14) analysis against the Swiss-Prot (15) database arbitrarily filtered at 1E−10. ESTScan was used with the derived model to predict the most likely peptide for each unigene sequence. The numbers of ESTs, unigenes and peptides are shown for each of the 55 openSputnik plant species along with estimates of actual coding potential and redundancy across the individual libraries (Table 1).

Table 1.

Table summarizing the sequence content of the openSputnik database

Organism name	No. of ESTs	EST sequence (bp)	No. of singletons	No. of assembies	Unigene sequence (bp)	Redundancy	Peptide sequence (aa)	Protein coding potential
Allium cepa	19 582	13 016 289	7252	4020	8 544 747	1.5	2 531 519	88.9
Arabidopsis thaliana	190 741	84 128 065	17 675	20 109	22 482 688	3.7	6 135 202	81.9
Beta vulgaris	20 151	10 184 665	9244	3706	7 368 791	1.4	2 015 990	82.1
Brassica napus	37 159	21 438 036	8041	5447	8 389 217	2.6	2 403 184	85.9
Capsicum annuum	22 433	10 226 020	7326	3056	5 496 951	1.9	1 477 080	80.6
Chlamydomonas reinhardtii	154 600	82 230 382	18 211	10 989	23 178 755	3.5	2 388 596	30.9
Citrus sinensis	23 337	12 738 998	5311	3416	5 474 795	2.3	1 473 294	80.7
Cryptomeria japonica	7128	3 624 193	3202	1203	2 457 784	1.5	579 834	70.8
Cycas rumphii	5952	2 873 079	2230	697	1 597 282	1.8	349 001	65.5
Eschscholzia californica	5468	2 529 150	3146	741	1 908 962	1.3	564 147	88.7
Glycine max	344 524	158 703 384	28 963	24 892	33 585 032	4.7	8 648 792	77.3
Gossypium arboreum	38 915	26 139 867	10 007	6076	13 043 919	2.0	2 958 835	68.1
Gossypium hirsutum	13 571	8 414 112	5934	1914	5 367 083	1.6	1 334 901	74.6
Hedyotis centranthoides	5416	2 476 009	3595	641	2 022 087	1.2	450 943	66.9
Hedyotis terminalis	4875	2 228 284	3313	530	1 830 094	1.2	402 306	65.9
Helianthus annuus	59 841	25 553 028	11 900	6050	8 654 947	3.0	2 086 806	72.3
Helianthus argophyllus	12 787	4 929 193	4646	1029	2 309 089	2.1	516 763	67.1
Helianthus paradoxus	10 340	4 149 627	3844	1012	1 997 115	2.1	458 465	68.9
Hordeum vulgare	372 431	198 114 717	25 405	23 033	37 345 565	5.3	9 139 515	73.4
Ipomoea nil	25 899	15 289 506	4572	4829	6 252 258	2.4	1 682 965	80.8
Lactuca sativa	68 188	35 969 889	12 427	7998	13 090 218	2.7	3 527 514	80.8
Lotus corniculatus	36 311	13 987 475	7646	4248	5 529 908	2.5	1 635 214	88.7
Lycopersicon esculentum	150 228	75 468 371	13 178	14 870	19 372 969	3.9	5 380 403	83.3
Lycopersicon pennellii	8346	3 842 358	2408	901	1 770 921	2.2	503 014	85.2
Medicago truncatula	187 763	101 662 463	19 448	17 189	27 597 708	3.7	6 630 342	72.1
Mesembryanthemum crystallinum	25 803	15 782 659	4831	3137	5 941 245	2.7	1 541 786	77.9
Nicotiana tabacum	10 323	5 104 499	8710	630	4 738 148	1.1	952 839	60.3
Oryza minuta	5268	2 367 832	2756	591	1 658 572	1.4	452 963	81.9
Oryza sativa	260 901	136 090 821	30 971	20 934	34 467 815	3.9	8 593 185	74.8
Phaeodactylum tricornutum	12 121	7 911 359	3043	1526	3 439 590	2.3	894 960	78.1
Phaseolus coccineus	20 120	8 487 980	4419	2431	3 269 096	2.6	886 986	81.4
Physcomitrella patens	102 219	54 477 833	10 114	13 309	15 177 696	3.6	3 521 525	69.6
Pinus pinaster	15 719	7 679 661	4974	2452	4 209 291	1.8	1 036 699	73.9
Pinus taeda	110 622	51 626 003	14 632	11 610	15 972 215	3.2	3 945 832	74.1
Poncirus trifoliata	6390	4 107 970	1644	1209	2 220 609	1.8	568 758	76.8
Populus alba	10 446	5 769 749	3856	1480	3 192 053	1.8	862 949	81.1
Populus balsamifera	30 296	14 140 412	7031	3664	5 503 910	2.6	1 522 330	83.0
Populus tremula	70 091	30 629 346	14 699	7954	11 475 126	2.7	3 192 054	83.5
Populus tremuloides	13 050	6 174 206	2634	2218	2 413 573	2.6	706 585	87.8
Porphyra yezoensis	20 979	9 801 783	2774	2045	2 853 651	3.4	681 731	71.7
Prunus persica	11 452	6 496 591	3206	1588	3 135 288	2.1	883 165	84.5
Saccharum officinarum	246 301	156 538 942	29 895	25 089	45 845 406	3.4	11 003 162	72.0
Saccharum spp.	8807	4 377 943	4784	1155	3 165 611	1.4	788 520	74.7
Secale cereale	9194	4 313 461	3793	1346	2 687 830	1.6	662 342	73.9
Solanum tuberosum	94 525	51 346 134	6651	15 983	16 752 895	3.1	4 715 299	84.4
Sorghum bicolor	161 766	83 411 684	16 955	17 704	23 132 774	3.6	6 004 630	77.9
Sorghum propinquum	21 387	9 750 610	5371	3507	4 673 286	2.1	1 209 822	77.7
Stevia rebaudiana	5548	3 242 045	2498	713	2 048 965	1.6	578 303	84.7
Theobroma cacao	6562	2 607 871	1988	753	1 103 776	2.4	276 188	75.1
Triticum aestivum	511 732	257 643 801	49 171	33 666	51 549 049	5.0	12 964 652	75.5
Triticum monococcum	9973	4 956 308	3941	1681	3 212 869	1.5	810 910	75.7
Vitis hybrid	6533	3 604 678	1032	1052	1 385 939	2.6	349 250	75.6
Vitis vinifera	135 712	74 769 503	9616	12 893	16 019 102	4.7	4 176 665	78.2
Zea mays	384 391	173 945 698	24 266	25 725	29 187 808	6.0	7 017 868	72.1
Zinnia elegans	9783	4 896 796	6536	1456	4 140 824	1.2	890 004	64.5

A total of 55 plant species are included in the current release, and represent a broad taxonomic distribution of species. Shown are the number of ESTs and the total nucleotide length for all EST sequences. The number of resulting singleton unigenes and multi-member assemblies is shown, along with the summed length of all available unigene sequence. The difference between total nucleotide length in EST and unigene sequences is summarized as apparent redundancy. Since peptide sequences have been prepared for each of the unigenes the length of all derived peptide is also shown and a measure of apparent coding potential across the whole unigene set is also shown.

DATABASE CONTENTS

The unigene sequences and peptides from each of the included species have been annotated using a selection of bioinformatics tools that are relevant to comparative genomics and biological understanding. Sequences are annotated for structural and functional characters using InterPro domains (16), TMHMM for the identification of transmembrane domains (17), TargetP for the prediction of organellar targeting (18) and SignalP for subcellular localization (19). The blast algorithm is used to reflect similarities of individual sequences with known proteins in the Swiss-Prot database, predicted proteins in the UniProt database (20) and to organism specific sets of proteins not restricted to A.thaliana, O.sativa or aggregated plant proteins. The complete sequence collections are summarized using the MIPS catalogue of functionally annotated proteins (Funcat) (21) and Gene Ontology terms (22). A collection of methods has been implemented to provide the typical figures and charts that are often seen in EST collection publications. Graphical representation of sequence lengths, number of ESTs within unigenes and clone-library representation are all included. Also included are reports summarizing the functional distribution of unigenes using both GOSlims and the MIPS Funcat.

DATABASE ACCESS

A query interface to the openSputnik database is provided by a web application product written for the Zope web application server. The openZputnik portal at http://sputnik.btk.fi provides access to all core EST collections through a single unified interface. Selecting EST projects will display a list of all available projects. When an openSputnik collection is selected, an interface that provides routes to the underlying data will be displayed. Different methods are included for EST sequences, unigene sequences and peptide sequences. Additionally, a page is included to access sequences on the basis of pre-computed reports and a BLAST server is included so that sequences may be identified on the basis of similarity to a known sequence. Sequences may be identified on the basis of a variety of criteria not restricted to GC content, length, name or predicted function.

When a sequence is selected, a single page summary report is displayed for the sequence. This summarizes key information that includes wherever appropriate, the best BLAST matches, functional information and physical attributes. Navigation tabs are provided so that a user may access all primary information derived or associated with a single sequence.

DATA AVAILABILITY AND FUTURE DIRECTIONS

All data within the openSputnik database is freely available to the scientific community. Please contact the author to request the inclusion of additional methods. The analytical pipeline may be applied to novel and proprietary sequence collections as either a collaboration with, or as a service of, the Bioinformatics Core facility provided at the Turku Centre for Biotechnology. The openSputnik SQL schema and complete database dumps are available upon request. The source code to the openSputnik engine and core reporting architecture is being open-sourced and released to Source Forge (www.sourceforge.com).

The openSputnik group will prepare one or two releases of the clustered plant unigenes per year. Additional plant species will be included into the pipeline as they exceed our arbitrary size threshold. Additional groups of organisms will be integrated in the future with a comparative mammalian unigene database planned for spring 2005. Additional emphasis is being placed on the creation of generic reports that can distil the essence of large and heterogeneous sequence collections. Further synchronization of the completed resources with the Gene Ontology and dynamic integration and comparison of groups of species is in progress. The challenge is to stay abreast with the ever-growing collections of sequences and the novel bioinformatics methodologies that offer us the ability to better understand the nuances within our sequence collections.