At maturity, C. rumphii leaves can reach up to 3 meters in length (Figure 1a). The tissue used in this study consisted of 10 to 40 cm of the immature leaf terminus protruding from the crown collected shortly after emergence (Figure 1b). Immature leaves consist of a petiole, a central rachis and circinate leaflets composed of both expanding and meristematic cells 30. RNA extracted from this tissue was used to construct a cDNA library from C. rumphii. Size fractionation was used to enrich for full-length cDNAs during library construction. It was determined that 53% of the cDNA clones were over 500 bp long. From this cDNA library, 4,210 sequence reads (ESTs) were generated. The majority of these reads (3,917) were generated from the 5' end of the cDNA; however, a small subgroup (293) were sequenced from the 3' end. Cluster analysis performed at the Munich Information Center for Protein Sequences (MIPS) of the entire EST dataset produced a UniGene set of 2,458 contigs consisting of 1,917 singletons and 541 assemblies. Of the clustered ESTs, the longest contig was 1,836 bp. The entire UniGene set can be viewed on the MIPS Sputnik website 31, which features sequence annotations and peptide sequence predictions. At the MIPS Sputnik site there are links to download the complete cycad sequences as an EST fasta file, a cluster fasta file or as the derived peptide fasta file.

Each contig from the database was automatically assigned to a functional category on the basis of its top match against the complete genomic sequence of Saccharomyces cerevisiae and A. thaliana databases using BLASTP. A non-stringent expect value (E-value) of <1e-10 was chosen as the threshold. The pie chart in Figure 3 illustrates the relative fraction that each functional category comprises within the entire UniGene set. The four largest predominant categories of cycad ESTs according to this functional categorization are: 'cellular organization' (22%), 'metabolism' (10%), 'unclassified proteins' (10%), and 'cell growth, cell division/DNA synthesis' (9%).

Using TBLASTX, a comparison was made between the C. rumphii UniGene set versus all available ESTs from GenBank and predicted Arabidopsis genes from The Arabidopsis Information Resource (TAIR). Both EST and predicted genes were grouped into three subcategories: angiosperms, gymnosperms, and lower plants. The angiosperm database encompasses all annotated rice and Arabidopsis genes identified from their respective genomic sequences, as well as all higher plant ESTs. The gymnosperm database contains ESTs from all gymnosperms, the majority of which came from the Pinus taeda EST sequencing project 3233. The lower plant databases included genes from all remaining plant ESTs including ferns, fern allies, bryophytes and algae available in GenBank. The angiosperm subgroup consisted of 84.5%, the gymnosperms 6.5% and lower plants 9.0% of the total genes used in this analysis.

The Venn diagram shown in Figure 4 displays the total number of cycad contigs shared between one or more of the plant gene datasets at very low BLAST stringency values (expect < 1e-5). The majority of cycad contigs (1,764/2,458) have counterparts in other plants, leaving 694 with no match to other plant genes. As one would expect, most Cycas hits (1,718) are to angiosperms, because of the predominance of angiosperm accessions in GenBank. Many of the cycad matches to angiosperms also match gymnosperms and/or lower plants (1,416). There are 1,310 cycad contigs that match gymnosperm genes and 734 that match genes from lower plants.

As shown in Figure 4, 44 Cycas ESTs specifically match only genes in the gymnosperm subgroup. Two additional Cycas ESTs match genes from gymnosperms and lower plants, but not angiosperms. To further analyze these 46 contigs that match only gymnosperms and/or lower plants, we next sequenced these Cycas cDNAs in their entirety to determine whether this 'gymnosperm/lower plant' specific grouping held up when the remaining portions of the cDNA were sequenced. Because ESTs, even when clustered into contigs, usually represent only a portion of the actual gene (particularly for genes poorly represented in the library) 37 of the 46 Cycas cDNAs were sequenced in their entirety (the remaining nine clones were not successfully recovered for sequencing), and this sequence can be downloaded from the Internet 34. Of these 37 fully sequenced cDNAs, 14 clones still showed no similarity to any known angiosperm genes, even at this low stringency cut-off. The insert size for each clone ranges from 586 bp to 1,899 bp, with predicted open reading frames (ORFs) varying from 69 to 527 residues (Table 1). None of these 14 Cycas cDNA clones is homologous to any known genes outside the plant kingdom, although Interpro analysis identified a small number of conserved motifs, which are listed in Table 1. To confirm that these genes were indeed derived from C. rumphii, gene-specific primers designed to each of the 14 genes were able to amplify a fragment from genomic DNA isolated from a different C. rumphii specimen and different tissue (sporophyll) from the source tissue of the cDNA library (data not shown). This distinct C. rumphii specimen was cultivated in a geographically separate location (Florida) from the cDNA source C. rumphii specimen used for cDNA library construction (New York).

A survey of the cycad EST dataset reveals a surprisingly large number of genes with highest similarity (BLASTP score < e-5) to genes with defined roles in growth and development in angiosperms (Table 2). Some of these Cycas genes have similarity to Arabidopsis transcription factors, including CONSTANS 3536, two distinct homeobox genes 37 and a YABBY gene 3839. Other cycad ESTs have similarity to other regulators of Arabidopsis development, including ARGONAUT 40 and COP9 4142.

A number of genes in our cycad EST library showed similarity to components of signaling pathways found in higher plants (Table 2). These genes include a photolyase blue-light receptor, genes involved in secondary signaling (including those for calmodulin, kinases, and phosphatases), a 14-3-3 protein, and genes involved in phytohormonal responses, including auxin (IAA-9 and IAA-13) pathways as reviewed in Chory and Wu 43. Surprisingly, a Cycas EST with high similarity to plant GLR-like genes was also found (Table 2) 1544. The presence of a GLR-like gene in cycads is of particular interest as it relates to BMAA, as described below.

BMAA, an agonist of mammalian GLRs, is a suspect causative agent of neurological disorders 913. However, nothing is known about the genes and enzymes involved in the biosynthesis of BMAA. Because the structure of BMAA is similar to other beta-substituted alanines 4546, it is likely that BMAA biosynthesis utilizes phosophoserine, cysteine, o-acetylserine or cyanoalanine as a beginning substrate. On this basis, a likely BMAA biosynthetic pathway is shown in Figure 5. This would require a two-step reaction initiated with the transfer of NH₃ at the beta-carbon of the substituted alanine (Figure 5a), followed by an addition of CH₃ (Figure 5b) to produce BMAA (Figure 5c). NH₃ transfer would require a nucleophilic reaction catalyzed by a cysteine synthase-like protein. A preliminary survey of genes in the cycad EST library identified candidate genes for both of these enzymatic steps (Table 2). The cycad leaf EST library contains two ESTs, which each encode a cysteine synthase. To catalyze the second step of BMAA synthesis, the EST library contains two potential methyltransferases (caffeic acid O-methyltransferase II and caffeoyl-CoA 3-O-methyltransferase). The second step would require a methyl donor, the most likely candidate being S-adenosylmethionine (SAdM). Consumption of SAdM would require the presence of enzymes to regenerate SAdM. A number of cycad ESTs can be implicated in SAdM recycling including: adenosylhomocysteinase, S-adenosylmethionine synthetase and homocysteine methyltransferase. Taken together, the cycad EST library contains candidate genes for all of the enzymes predicted to be present during the biosynthesis of BMAA.

Extant genera, such as Cycas, have changed little in morphology from their extinct relatives, such as Crossozamia, which existed during the Permian 12. The study of cycads has proved to be useful in reconstructing plant evolution, in particular in understanding the rise of important plant structural innovations such as the evolution of seeds 47. Cycads also produce a variety of neuroactive compounds, some of which are suspected to be the source of Guam's dementia 1148. However, despite their scientific importance in plant biology and medicine, virtually nothing is known regarding gene expression, development and signaling in the Cycadales. As a first step in this direction, a cDNA library was made from young, developing C. rumphii leaves to produce a cycad EST database.

One advantage of a genomics approach is that it provides rapid access to genes important for evolutionary studies. The more traditional homology-based gene-cloning approach is limited by tedious gene-by-gene purification. It is also limited in that it may miss related genes if the degeneracy is too great or if nonconserved regions of the protein are chosen during primer design. Finally, the targeted gene approach can never be used to discover new genes.

An EST project in Pinus taeda (loblolly pine) sampled 59,797 transcripts from wood-forming tissues 32. In this analysis, 66 P. taeda contigs showed BLAST similarity at low stringency only to other gymnosperms. Similarly, in our analysis, we found 46 cycad contigs that only matched gymnosperms (including P. taeda) and/or lower plant ESTs, but were not found in the genomes of higher plants or non-plants. Complete sequencing of 37 of these cycad cDNA clones showed that 14 clones, ranging in length from 586 to 1,899 bp, were still found only in other gymnosperms. Having no homology to the completely sequenced genomes of two different angiosperm species - Arabidopsis 19 (a dicot) and rice 2324 (a monocot) - suggests that these 14 genes are found only in gymnosperms or lower plants, in which genomic studies have only just begun. However, because ESTs as well as contigs usually represent only a portion of the full-length gene sequence, these results are preliminary. For instance, in P. taeda, larger contigs have a higher BLAST match rate to other plant genes then do shorter contigs 32. Thus, these preliminary results of clade specificity are tenuous and presumably will change as more ESTs, as well as full-length gene sequences, from cycads and other species are generated in the future.

As in higher plants, cycad leaves are derived from the shoot apical meristem (SAM) 30. In Cycas leaflet primordia, meristematic growth ceases at the apex, while proceeding basipetally where it becomes localized to the leaflet margins 30. The presence of these marginal meristems may explain why a surprising number of developmental genes were identified in a relatively small number of ESTs from young cycad leaves (Table 2).

A gene with identity to the YABBY gene family was among the cycad ESTs. YABBY genes encode transcription factors expressed on the abaxial side of all lateral organs that promote abaxial cell fate 38. In Arabidopsis, mutations in the YABBY gene INO (INNER-NO-OUTER), lead to the loss of the outer integument 49 reminiscent of gymnosperm (and cycad) unitegmy (the presence of a single integument). Unitegmy is considered to be the ancestral condition in seed plants 547. An analysis of YABBY gene expression in cycads may help to explain the origin of the integument in gymnosperms, and/or possibly the second integument in angiosperms. One cycad EST from the library has highest similarity to COP9. COP9 encodes a subunit of the COP9 signalosome complex, which controls multiple signaling pathways that regulate development in all eukaryotes 4250. In Arabidopsis, the cop9 mutant is constitutively photomorphogenic in dark-grown seedlings 51. Some gymnosperms, (in particular the Coniferales) are constitutively photomorphogenic when grown in the dark 5253. As yet, the phenotype of dark-grown cycad seedlings has not been fully evaluated. The discovery of a gene encoding a putative subunit of the COP9 complex in cycads could be a first step to define the ancestral, developmental role of the signalosome in gymnosperms, particularly with regard to its role in photomorphogenesis.

Another gene potentially involved in cycad development has highest similarity to the CONSTANS gene family, which are regulators of flowering time that follow internal and external (environmental) inputs in Arabidopsis 35. Because cycads predate the evolution of flowers, it would be of interest to determine if CONSTANS genes in cycads temporally regulate sporophyll and cone induction, which typically follows a yearly cycle 56.

An unexpected finding of the Arabidopsis EST genome project was the discovery of GLR-like genes, or 'neural' receptor genes, in plants 15. In Arabidopsis, the GLR-like gene family comprises 20 members 54. Pharmacological evidence has linked Arabidopsis GLRs to light and/or growth signaling pathways 1516. Supplying exogenous BMAA to growing Arabidopsis seedlings was shown to block light-induced hypocotyl shortening and cotyledon expansion 16. Because BMAA has such profound effects on Arabidopsis development, we have previously proposed that BMAA, or glutamate, the natural agonist of GLRs in humans, plays a physiological role in Arabidopsis 1516. Continuing genetic studies in Arabidopsis aim to identify the endogenous components of the BMAA-targeted pathway in plants 16.

Cycads produce BMAA 89. One EST uncovered in the C. rumphii leaf cDNA library has a high degree of similarity to plant GLR genes (Table 2). This discovery is intriguing, because it suggests that BMAA might be interacting with native GLR gene products in cycads. To further investigate the relationship between cycad GLR genes and BMAA, we sought to identify cycad genes potentially involved in BMAA synthesis.

From the structure of BMAA, we hypothesized that cycads produce BMAA in a simple two-step pathway, beginning with a β-substituted alanine. To enhance the probability of finding genes involved in BMAA synthesis, we made our cDNA library from tissues that produce relatively large quantities of BMAA (nearly 0.1 mg/g tissue) 28. According to Ohlrogge and Benning, there is a 95% chance of finding the gene for a specified enzyme when it is expressed at 0.1% mRNA/protein by sampling only 3,000 ESTs from an unnormalized library 55. Considering the prevalence of BMAA in Cycas, it is not surprising that we discovered cognate genes for the predicted enzymes for this BMAA biosynthetic pathway in the cycad EST database (Figure 5, Table 2). Future biochemical and molecular studies will determine if these genes play a part in BMAA synthesis.

The discovery of GLR-like genes in C. rumphii raises the intriguing possibility that endogenous BMAA may interact with native cycad GLRs as a regulatory molecule. Future studies aim to understand the role of GLRs in plants, as well as the role of BMAA in herbivore defense versus endogenous signaling. The production of additional ESTs from cycads will increase the variety of genes available for study, so that a detailed expression profile can be evaluated during cycad development. Complementation studies of these genes in orthologous Arabidopsis mutations will help define their roles in cycads. This combined approach to studying cycad gene structure and function will help reveal molecular changes in genes involved in signaling, metabolic and developmental pathways that led to the rise of the seed plants.

Newly emerged immature leaves from the crown of a C. rumphii tree, accession 808/59 A, were collected from the New York Botanical Garden Conservatory. Leaves collected ranged from 5 to 30 cm in length. Tissue was frozen in liquid nitrogen. RNA was extracted from pulverized, frozen tissue in a mortar and pestle with the RNeasy maxi kit (Qiagen, Valencia, CA) according to the manufacturer's protocol. Purified Cycas RNA was precipitated in 2 M LiCl, washed twice with 70% ethanol, and resuspended in 50 μl water. Poly(A) RNA was subsequently purified from total RNA with the Oligotex Maxi kit (Qiagen). A cDNA library was constructed using the Lambda ZAP-CMV cDNA synthesis kit (Stratagene, La Jolla, CA) using 10 μg poly(A) RNA. Before cloning, cDNA was size fractionated over a Sepharose CL-6b column. The first five fractions containing a total of around 100 ng cDNA were collected, pooled and precipitated in 70% ethanol/0.3 M sodium acetate and resuspended in 3.5 μl water. cDNA (0.5 μl) was then directionally subcloned into the vector at the EcoRI and XhoI sites.

DNA was collected from unemerged C. rumphii sporophylls using the DNeasy purification kit (Qiagen).

Plasmid DNA was collected as described in the manual (Stratagene) catalog number 200450 in the in vivo mass excision section. Sequence analysis was performed at Cold Spring Harbor Laboratory using an ABI 3700 capillary sequencer (Applied Biosystems, Foster City, CA) for separation and nucleotide detection. Reactions were performed using a 1/16 Big Dye Terminator. Sequencing was performed with either the -21 M13 forward and/or reverse primer.

The EST sequences were clustered and assembled using the HarvESTer application (Biomax informatics, Martinsried, Germany). The default HarvESTer settings were optimized to screen for vector against the UniVec nonredundant database of vector and polylinker sequences 56. The Hashed Position Tree (HPT) clustering used a similarity link threshold of 0.7 and a maximum distance of six steps was required to define a cluster from the similarity network, thus encouraging the separation of likely paralogs. Cluster consensus sequences and concomitant alignments were derived from the HPT clusters using the CAP3 application with default settings. The HarvESTer assemblies and coordinate alignments were imported into the Sputnik EST and cluster analysis application 57.

BLASTX 58 was performed against a nonredundant protein database for each of the cluster consensus sequences. Likely coding sequences were derived for each cluster consensus sequence by parsing the best BLASTX match and filtering the results using the arbitrary expect value <1e-10. Dicodon usage frequencies and probabilities were extracted using tools from the ESTate package 59. A peptide sequence was predicted for each of the cluster consensus sequences using the Framefinder application from the ESTate package with the cycad-specific codon usage statistics. Framefinder was run using the default parameters. The derived peptide sequences were used as the basic scaffold for peptide-based annotation in Sputnik.

Sequence annotation on each of the cycad cluster consensus sequences and derived peptides were performed within the Sputnik application. Results were assessed for possible contamination by searching for homology to the Escherichia coli and human genomes and were scored for homology to a wide range of noncoding RNAs and plant chloroplast and mitochondrial genomes. Similarity searches were performed using the BLAST application 58 and results were filtered using the expectation value < 1e-10. Functional assignment was performed on both cluster consensus sequence and the peptide sequence. Assignments were made using BLASTX and BLASTP respectively against the MIPS catalog of functionally assigned proteins (funcat) 6061: tentative functional assignments were filtered using the expectation value < 1e-10.

All cycad contigs sequences were aligned against the PlantEST database using TblastX 58 and BlastX against the NR(aa) database. The PlantEST database was created by downloading all plant ESTs in GenBank and assembling them using Phrap 6061. Todd Wood from Clemson University provided the PERL script that creates the PlantEST databases as described above. The NR(aa) database is a nonredundant database of protein sequences from GenBank.

All available plant ESTs were downloaded from GenBank and separated into three datasets consisting of angiosperms (monocots and dicots), gymnosperms, or lower plants (ferns, mosses and algae). Downloaded ESTs were assembled using Phrap 6061. All matches with an expect value < 1e-5 were considered significant.