Utilizing the assembly algorithms incorporated in the SeqManII software (part of the DNASTAR suite from DNASTAR Inc, Madison WI), the 9,159 ESTs were grouped into contigs and clusters by establishing assembly stringencies that generated groupings defined by the analysis protocols of McCarter et al., 2003 34. Contigs contain EST members that appear to originate from single transcripts whereas clusters are assemblies of ESTs that could represent transcripts from the same gene but alternate splice isoforms, or in the case of Tortula ESTs, which derive from a population of individual gametophytes, alleles of the same gene. The 9,159 ESTs formed 7,272 contigs and 5,563 clusters, both of which exhibit an average size of 669 nucleotides. However, the longest sequence increased from 1,575 nucleotides for contigs to 2,229 nucleotides for clusters. Clusters varied in content from a single EST (singletons) in 4,362 cases (78%) to 48 ESTs for a single cluster (Figure 1). The elimination of redundancy during contig building and cluster formation reduced the total number of nucleotides for further analyses from 5.93 million to 4.87 million (contigs) and 3.71 million (clusters). Overall the 9,159 ESTs potentially represent 5,563 genes, a discovery rate of 60%, with 47.6% of the ESTs as singletons. This is an overestimation of gene discovery since several non-overlapping clusters can represent a single gene and from our blast search data this appears to be a possibility, at least in the case of clusters 121 and 204 that appear to independently represent a gene that has a weak similarity to the LEA protein of Caenorhabditis elegans. Even though the ESTs derive from a non-normalized library 96.5% of the clusters still have 5 or fewer EST members.

The consensus Cluster sequences were subjected to a BLASTx style search of a custom curated non-redundant database derived from UNIPROT and annotated according to the degree of similarity of the cluster to the highest scoring match in the database of known gene sequences. The degree of similarity was based on the quality of the BLASTx statistical outputs as well as a visual inspection of the aligned sequences between the query and the target. Clusters that generated an HSP-bit score below 70 and or E values higher than 10^-7 were assessed manually for all possible alignments taking into account the alignment length, number of identical matches, gaps, and positive replacements. Utilizing these criteria we were able to annotate 3,321 (59.7 %) of the 5,563 clusters by their similarity to known genes within the database. This also meant that 2,242 of the clusters, or 40.3%, represent sequences that have no known counterpart in the public databases and that we categorize as unknowns.

Table 1 lists the 30 most abundant EST clusters derived from the Tortula rehydration EST collection. These transcripts only account for 8.4% of the generated ESTs, however. Ten of the most abundantly represented transcripts encode proteins that do not match any of the sequences in the databases searched in this study and are designated as unknowns. Seven abundant transcripts appear to encode proteins that belong to a class of proteins known as Late Embryogenesis Abundant (LEA) proteins, although the relatively low HSP bit scores and high E-values for most of these BLASTx matches has to be considered as a caveat in this assessment. LEA proteins have long been ascribed a protective role for cells that are experiencing dehydration 1135. This is a conclusion drawn on the strong correlation between LEA transcript accumulation and water loss, rapid decline in transcript levels upon rehydration, and especially as LEA gene expression relates to the programmed desiccation stage of seed maturation. If one can conclude that LEA protein transcripts are abundant in the rehydration transcriptome of Tortula ruralis, which is consistent with some of our earlier findings 32, then it is also possible that these proteins are involved in either protection of cellular integrity during the initial phases following rehydration when cell disruption is apparent in bryophytes 36, or are actively involved in the restoration of cells damaged by a desiccation event. This difference in the response of LEA gene expression between Tortula and what has been reported for angiosperms may also be a reflection of what we believe to be a more primitive mechanism of dehydration tolerance, and perhaps a more primitive role for and control of LEA gene expression. Other abundant transcripts appear to fall into the membrane transport (aquaporin, cysteine rich proteins, channel and pore proteins) and proteins that can be associated with plant stress events (metallothionine, esterase, and rubredoxin). The Early Light Inducible Protein A (ELIP-A) transcript is the only member of the abundant transcripts that we have previously reported 33 as belonging to the group of proteins we have termed, rehydrins 31. These proteins have been suggested to be synthesized in response to stress-induced photo-damage within the Tortula chloroplast and may play a protective or repair function for the photosynthetic apparatus 37.

Physiological, biochemical, and molecular data all point towards an active period of cellular activity, presumably for repair and recovery from desiccation induced damage, during the first two hours following rehydration of dried gametophytes 1315. The identity of the more abundant transcripts (Table 1.) does provide some insight into the nature of the metabolic activity that is associated with rehydration, at least it gives an indication of what metabolic processes may be of prime concern as the plant recovers from desiccation. However, the functional classification of the Tortula transcripts present during the initial phases following rehydration using the Gene Ontology (GO) classification system paints a broader view of the possible metabolic activity of the gametophytic cells at this time. This does, of course, come with the understanding that these are transcript based analyses and do not directly reflect protein levels which would offer a more definitive assessment of the metabolic capability of the cells during rehydration.

Functional classification of the Tortula rehydration transcripts was achieved by matching the Tortula clusters to characterized protein domains in a combined protein database, using HT-GO-GAT (Materials and Methods), which allowed us to assign GO terms to each cluster. The assignment of GO terms made it possible to place the clusters into the GO hierarchy which can be viewed by use of an AmiGo browser. Of the 3, 321 clusters that exhibited significant similarity to known genes in the public databases, 2,203 (66% of annotated or 40% of all clusters) represent genes that contain conserved protein domains that have known biochemical and physiological functions in other organisms and map to the GO hierarchy. The GO representations for the Tortula rehydration clusters are presented in Tables 2 through 4. The representations are segregated into the three main organizing principles of GO: biological process (Table 2), cellular component (Table 3), and molecular function (Table 4). A complete listing of the GO mappings is available on our website at http://199.133.147.108/Bryobase.

The representation of the Tortula rehydration clusters throughout the GO mapping system is indicative of the emphases on, but not expression levels of, particular cellular activities represented in the moss gametophytes during this period of a wet and dry cycle. In an attempt to assess if the accents on particular cellular activities indicated for rehydrated gametophytes are characteristic of the rehydration induced metabolic state or are simply indicative of processes associated with normally active bryophyte cells, we compared the representation of the Tortula rehydration clusters within the GO categories with similar "clusters" from Physcomitrella patens, the only other bryophyte that has similar genomic level information. The majority of the Physcomitrella "clusters" are derived from a large EST collection representing transcripts from both untreated and hormone induced (to switch developmental pathways) cells of protonemal cultures. The Physcomitrella ESTs are described by Nishiyama et al., 38 and were obtained from Physcobase http://moss.nibb.ac.jp/ as assembled contigs (assembled in an identical fashion to what we designate as clusters). In total 22,885 Physcomitrella contigs, derived from 102553 ESTs obtained from Physcobase and Genbank, were subjected to a BLASTx search, as described for the Tortula clusters using HT-GO-GAT. Of the 22,885 contigs 9,981 (43.6%) represent genes that contain conserved protein domains that have known biochemical and physiological functions in other organisms and map to the GO hierarchy.

There are two caveats for this comparison; 1) differences in representation may simply reflect species differences in the emphasis on individual classes of cellular activities between Tortula and Physcomitrella, or 2) differences in representation may reflect differences in the emphasis on individual classes of cellular activities between mature gametophytes (Tortula) and protonema (Physcomitrella). Until a comparison of GO mapping distributions can be made directly between clusters derived from an EST collection from hydrated control gametophytes from Tortula with those from the rehydration collection these caveats remain important. Nevertheless, even with these difficulties and limitations the comparison is still useful for developing new hypotheses as to what cellular processes might be important in the recovery of moss cells from desiccation.

In general the distribution of the Tortula clusters within the GO mappings are similar if not identical to the distribution of the Phycomitrella contigs gleaned from Physcobase. This can be seen especially at the GO map 2 and 3 relationship-level categories (Table 1 and supplemental material) using the multi-species GO browser (DrZOOview2.0 http://199.133.147.108/mainbioinformatics.html). However, there are several notable differences and in general the differences are consistent with what has been determined, in earlier studies, to be important in the recovery of the moss from desiccation following rehydration 423 and what is known about plant responses to abiotic stress in general 813. The differences evident in the comparison are presented in Table 5, where the extent of the similarity in the distribution of the clusters are expressed as the ratio of representation for Tortula to representation for Physcomitrella (T:P). In the Biological category the most striking differences in GO representation of Tortula clusters occurs in categories where the percentage representations are relatively low but the numbers of clusters are substantial. In the level 4 relationship, Photosynthesis, the representation for Tortula is 3.31% compared to 1.38% for Physcomitrella, a 2.4 fold difference, the majority of which is accounted for by the difference in the representation levels for the light reaction category. The rapid recovery of photosynthesis is critical for the recovery of bryophyte cells, particularly in regards to the production of energy and reducing power for the metabolic activity associated with repair and reconstitution of the gametophytic cells. Chloroplast structure in Tortula is severely disrupted, especially if desiccation occurred rapidly, in the first few hours following rehydration 3640 but recovers quickly. The difference in representation in this category is consistent with these observations and thus may reflect the greater need for a supply of a diversity of photosynthetic components in rehydrated Tortula than in Physcomitrella cells that have not experienced a disruption in the photosynthetic apparatus. Similar inferences can be made concerning the differences in representation observed for carbon utilization (T:P of 2.0) and oxidative phosphorylation (T:P of 2.1) for Tortula in that mitochondrial activity and integrity are also compromised during rehydration 2536. Other differences in representation between Tortula and Physcomitrella mappings relate to a higher representation for Tortula in the categories that relate to responses to external stimuli and responses to stress both of which would seem consistent with the emphasis that cellular activity for Tortula would have in comparison to unstressed Physcomitrella cells. In particular the increased representation within the response to oxidative stress is of note, as an elevated protection of cellular integrity from the damaging reactive oxygen species (ROS) typically associated with a desiccation rehydration event is a distinctive component of desiccation tolerant bryophytes when compared to their desiccation sensitive relatives such as Physcomitrella 41.

The above observations concerning the more extensive representation of clusters within the GO mappings related to organelle function in Tortula are mirrored in the Cellular Component Level 2 classification mappings. In the Cellular Component classification Tortula exhibits an almost two fold difference in representation within categories associated with either the chloroplast or mitochondria, such as Extrachromsomal DNA (T:P of 1.8), Thylakoid (T:P of 1.8), Mitochondrial membrane (T:P of 1.9), and both Inner and Outer membranes (T:P of 1.8). In this case the categories are generally related to genes representing membrane components and since it is the organelle membranes that exhibit the majority of the damage during desiccation and rehydration it is consistent that these categories would be better represented in the Tortula cluster mappings than those for Physcomitrella. In addition to the organellar related classifications the Tortula clusters also exhibit a higher representation within the Ribonucleoprotein complex category (T:P of 1.8) which in all likelihood reflects an emphasis on ribosomal components since a similar difference in representation is seen in the Structural Constituent of the Ribosome category (T:P of 2.0) within the Molecular Function Level 2 classification. Again such differences in representation in the comparison between Tortula and Physcomitrella GO mappings are consistent with our previous studies on the responses of Tortula gametophytes to desiccation and rehydration and comparisons to non-stressed bryophyte tissues. Protein synthesis is critical to the recovery of Tortula cells following a desiccation event 111315 not only for the synthesis of proteins damaged by the stress of desiccation but also directing the response to the stress at the level of gene expression 2629. Early studies determined that the speed at which desiccation occurred has a marked effect on both the rate of recovery of protein synthesis and the rate at which either new ribosomes are formed or pre-existing ones are repaired 4243, rapid desiccation results in a more prolonged recovery of normal protein synthetic levels and also slows the reconstitution of ribosomes upon rehydration. Since the Tortula clusters are derived from ESTs of rehydrated moss that was dried rapidly, the greater representation in ribosome related GO mappings for this collection compared to the Physcomitrella clusters, that represent transcripts from cells where presumably normal ribosomal turnover and synthetic rates are prevalent, is consistent with the biological state of the Tortula cells. The emphasis on protein synthesis in rehydrated Tortula cells compared to those of Physcomitrella is also evident in the comparison of representations within the Translation regulator, Translation factor, and Translation elongation factor activity GO mappings.

Other differences seen in the Molecular function classification, such as the greater representation within the Tortula collection of clusters involved in Two-component Sensor or Channel/pore class Transporter activity designated mappings, offer novel possibilities for investigation into rehydration metabolism that have not been indicated as important until now. The individual identity of the clusters that map to these categories should offer possible hypotheses that can be tested in our future research.

An alternative classification of clusters, based on biochemical function, involves the use of HT-GO-GAT to assign clusters to individual Kyoto Encyclopedia of Genes and Genomes (KEGG http://www.genome.ad.jp/kegg/kegg2.html) metabolic pathways. Of the 2203 clusters that map within the GO hierarchy only 642 clusters had assigned EC numbers generating 325 unique mappings (see additional file 1: KEGG biochemical pathway mappings for Tortula rehydration clusters). The paucity of EC assignments limits this aspect of the analysis but the mapping of clusters to the KEGG metabolic pathways still presents some useful perspectives on the metabolic emphasis of the rehydrated gametophytic cells. Eighty-six of the 123 pathways contained within the Metabolism category (metabolic pathways), were represented by 84.6% of the 642 Tortula clusters. The KEGG metabolic pathways that are well represented by Tortula clusters are Carbohydrate Metabolism (83 enzymes represented), Amino Acid Metabolism (72 enzymes), Energy Metabolism (40 enzymes), Lipid Metabolism (27 enzymes), and Metabolism of Cofactors (24 enzymes). All of these pathways have been previously associated with the cellular recovery processes associated with rehydrated moss gametophytes 1415. Within the metabolic activities not represented by the Tortula clusters only the lack of ascorbate metabolizing enzymes appears unusual as ascorbate has been well documented as an important metabolite in the protection of moss gametophytes from oxidative damage during stress. This however appears to be the result of a limitation in the assignment of EC numbers since several Tortula clusters show significant similarities to enzymes involved in ascorbate metabolism in the original BLASTx search used to generate the GO mappings. The limitation of the KEGG based classification can also been seen in the poor representation of Tortula clusters in the other KEGG pathways (Genetic Information Processing, Environmental Information processing, and Cellular Processing) which is surprising given the number of clusters that were identified in the BLASTx search and in the GO databases as being contained within these classifications. As an example, none of the 239 clusters that map in the GO hierarchy as structural components of the ribosome (Table 4) or any of the clusters that mapped to transcriptional and translational components were contained in the corresponding KEGG pathways. Thus although useful information can be gleaned from the KEGG classification system and metabolic pathway mappings, especially in practical terms for functional studies using individual clusters, it has some major limitations for drawing any broad based hypotheses from the representation of Tortula clusters within each pathway.

Of the 5,563 consensus cluster sequences used in the BLASTx search of our database (see above) 40.3% failed to exhibit sufficient similarity (did not meet set criteria, see above) with known sequences to allow for an accurate annotation of the cluster. It is possible that these contigs, rather than containing novel amino-acid coding regions, contain mainly 3' or 5' untranslated regions (UTRs) or coding regions that are so short as to render them incapable of generating a significant similarity score. In order to investigate this possibility we examined the three classes of contigs, those with significant similarity scores; "good hits", those that generated poor similarity scores; "false hits", and those that failed to generate any scored similarity; "no hits", to determine the longest open reading frame (ORF). We limited the ORF determination to those clusters that contain an AUG codon in the 5' to 3' direction of the clone in any one of the three possible reading frames (the cDNAs were directionally cloned). The results of this analysis are shown in Figure 2. Of the 5,563 clusters generated in the study, 4,789 generated ORFs under the limitations imposed by the analysis. Of these 2,983 were classified as "good hits", 1,564 as "false hits" and 242 as "no hits". Those clusters that are classified as "good hits" exhibit ORFs that are in general evenly distributed from 20–40 amino acids long to 220–240 amino acids long. The clusters that are classified as "false hits" from the BLASTx search do have a relatively larger proportion of shorter ORFs, in the 20–40 and 40–60 amino acid range but also a substantial proportion that are much longer. The distribution of ORFs in this category ('false hits") does not appear to be sufficiently skewed from that for the "good hit" clusters to render them incapable of generating similarity scores in the BLASTx search. This would suggest that these clusters do contain novel amino acid sequences that are not represented in the public protein databases by sequences sufficiently similar to generate significant HSP Bit scores. In addition, the distribution of ORFs in the "no hit" cluster category are distributed in a similar manner to those of the "good hit" classification and so are also likely to represent clusters encoding proteins with novel amino acid sequences. The clusters contained within the "false hit" and "no hit" categories are of particular interest in our search for novel genes and pathways that are associated with the ability of certain plants to acquire vegetative desiccation tolerance.

In this analysis the Physcomitrella and Arabidopsis databases were independently used in a BLASTx search using the 5,563 Tortula clusters as query sequences. Even though we used the entire 5,563 cluster sequences as individual queries only 3,321 actually represent sequences that could be annotated using the criteria for similarity discussed previously. When the Tortula clusters were used as queries against the Physcomitrella database, 5,554 generated HSP Bit scores however only 282 exhibited a level of conservation of sequence that passed the criteria established for annotation of our Tortula clusters (HSP Bits score above 70 and or E values of 10^-7 or less, see above). Against the Arabidopsis database only 927 Tortula clusters generated HSP Bit scores but 612 had a level of conservation sufficient to be considered reliable matches to Arabidopsis genes. These observations are difficult to rationalize and may simply reflect the inequality of the target databases. The Physcomitrella database, which is generated from a relatively small EST collection, is much smaller than that for Arabidopsis, which is gleaned from the full genomic sequence. This may explain why although most Tortula clusters generated HSP Bit scores only a relative few were capable of producing scores sufficiently high enough, and with low enough error probabilities, for confident assignment of co-identity with a Physcomitrella contig. Obviously Tortula ruralis is more closely related to Physcomitrella patens, as they are both bryophytes, than to Arabidopsis thaliana but the phylogenetic distance between the two bryophytes is still substantial, which may also help explain the paucity of high value matches and the large number of low value hits. The Arabidopsis database is much more comprehensive and would be expected to generate more significant matches as observed and perhaps because of the evolutionary distance between the two plants more queries that do not generate an HSP.

The twenty gene products that exhibit the highest level of conservation for both comparisons, Physcomitrella (E values of 0 to -39) and Arabidopsis (E values of 0 to -55) are presented in Table 6. Several common highly conserved genes are present in both comparisons, including genes involved in cell structure (tubulin and actin), protein synthesis (ribosomal proteins and elongation factors), protein-turnover (polyubiquitins), stress proteins (heat shock), chromosomal proteins (histones), signal transduction (ADP-ribosylation factor), and binding proteins (calmodulin). Interestingly there are differences between the two lists, which may reflect the relative phylogenetic distances between the three species. The three most conserved proteins between Tortula and Physcomitrella did not register as highly conserved between Tortula and Arabidopsis, the most conserved protein between Tortula and Physcomitrella, the rRNA intron-encoded homing endonuclease, did not generate a match at all with an Arabidopsis counterpart. Interestingly, the most conserved photosynthesis related protein between Tortula and Physcomitrella is a chlorophyll a/b-binding protein which generates a HSP Bit score of 205 and E value of 3.0E^-53, the Arabidopsis counterpart generates a HSP Bit score of 82 and E value of 8.9E-15. In this case the level of conservation of this protein within the comparisons appears to reflect the phylogenetic relationships that exist between the three plants. However, the most conserved photosynthesis related protein between Tortula and Arabidopsis is a photosystem I P700 apoprotein A2 which generates a HSP Bit score of 1120 and E value of 0, the Physcomitrella counterpart generates a HSP Bit score of 34 and E value of 1.25E^-1, which appears contrary to what is expected at what is seen for the chlorophyll a/b-binding protein. This type of relationship is also seen for the Ribulose 1 protein (Cluster 2170) that is conserved between Tortula and Arabidopsis but only to a low degree with the Physcomitrella counterpart. If such relationships can be substantiated with further study and complete sequences (both the Tortula and Physcomitrella clusters are based on ESTs) then some interesting evolutionary questions can be raised concerning both gene function and selection pressures on the individual plants and metabolic processes.

Tortula ruralis ([Hedw.] Gaertn, Meyer and Scherb), also classified as Syntrichia ruralis, gemetophytes were collected, harvested, and stored as described previously 27. For experimental purposes, gametophyte tissue was hydrated for 48 h to fully recover from dried storage and trimmed to remove stem material. Rapid-dried moss was prepared by placing the cropped gametophytes in a closed atmosphere of 0% relative humidity (RH) on 3-mm filter paper over activated silica gel in a Petri dish. This drying regime resulted in the attainment of the air-dried state within 30 min. The gametophytes remained in this atmosphere overnight to ensure desiccation and prior to library construction were rehydrated for 2 h in deionized water at 18°C in the light.

Total RNA was isolated from the rehydrated gametophytes by a series of phenol extractions as described by Lane and Tumiatis Kennedy 47. PolyA RNA was isolated from the total RNA fraction by oligo-(dT) chromatography 48, using DynaBeads oligo-(dT25) (Dynal, Inc., Lake Success, NY, USA), through two rounds of selection according to the manufacturer's instruction. The purified polyA fraction was used as a template for double-stranded cDNA synthesis using the Superscript Plasmid System (Invitrogen Life Sciences, Carlsbad, CA, USA). The resultant cDNA population was cloned into the pSPORT1 vector, according to manufacturer's instructions, to construct the unidirectional rehydration cDNA library. Small scale sequencing of 384 random clones confirmed the directional aspect of the inserts, the plant nature of the source cDNAs, and the frequency of positive clones. The average insert length for the library was assessed at 1.2 Kb. A subset of 10, 368 randomly picked positive clones (white in a blue-white X-Gal/IPTG based screen) were transferred to individual wells in 384 well plates containing suitable growth medium for storage, replication, and sequencing.

High throughput sequencing of the inserts contained in the 10,368 individual clones was performed using "rolling circle amplification" of the individual plasmids to generate suitable sequencing templates at the Joint Genome Institute, Walnut Creek, CA, U.S.A. Clones were sequenced using primers specific for vector sequence upstream of the multiple cloning site and at the 5' end of the cDNA insert. The sequences were delivered as primary binary files (raw trace files), which were then processed through our sequencing, pipeline as described below.

Trace files were entered into SeqMan II and quality screening performed with medium stringency corresponding to a phred threshold value of 12. Vector searching was performed using the pSPORT™ vector both in forward and reverse orientations with minimum match length of 7, connect distance of 3, Maximum register shift of 10, Minimum NW percent match of 90, gap weight of 0 and length weight of 2. Contaminant screening was performed with minimum match of 25. Following the quality screening the sequences were exported as a single FASTA file and polyA tails were removed using a custom PERL script. Several hundred sequences that contained internal polyA stretches were manually identified. These sequences were manually edited to remove polyA and trailing sequences.

Edited sequences were then entered into SeqMan for assembly into contigs. Assembly for contigs was performed with match size of 50, minimum match percentage of 97, minimum sequence length of 100.

Consensus contig sequences were exported from SeqMan as individual files and entered into a separate SeqMan project and reassembled using the same parameters used for contigs but with a minimum match percentage of 90. These were considered clusters.

EST submission to GenBank was performed using the USDA-ARS Livestock Issues Research Unit's (LIRU) High Throughput-Gene Ontology-Genome Annotation Toolkit (HT-GO-GAT). HT-GO-GAT can be obtained from the LIRU website at http://199.133.147.108. A total of 9159 EST sequences were submitted to GenBank and assigned accession numbers CN200321-CN209479

Functional annotations, Enzyme commission numbers, and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway assignments were assigned to cluster sequences using HT-GO-GAT. Consensus sequences derived from Clusters were entered into the software that utilizes custom BLASTx, RPS-BLAST, and relational mySQL databases to identify potential functional assignments based upon sequence and functional domain similarity matching. The software was set to identify high stringency matches and the resulting data manually curated using the high throughput results viewer interface of HT-GO-GAT. HT-GO-GAT produces various reports related to given datasets including ORF analysis, KEGG pathway reports and visualization, and also exports gene ontology association files. These association files were imported into a mySQL database and visualized using the USDA-ARS-LIRU's DrZooView2.0 GO database browser http://199.133.147.108.