A total of 6,448 cDNA clones were bi-directionally sequenced, resulting in 12,896 raw chromatograms, which served as input for the processing pipeline. After base calling by phred 37, Lucy 38 discarded 2,556 low-quality sequences, short or insert-less sequences, and vector or polyA-only sequences. An additional 55 sequences were removed by seqclean 39, leaving 10,285 high-quality ESTs (from 5,450 cDNA clones) for further processing (success rate ~80%). Assembly of these ESTs by cap3 40 resulted in the generation of 1,427 contigs, which ranged from 112 to 3,440 bp in length and contained 2 – 259 ESTs (mean: 4.8). Together with the remaining 3,498 singletons, a total of 4,925 unique sequences (UniSeqs) were generated. Because of the possibility that two (or more) UniSeqs originated from the same transcript, we also estimated the number of unique genes (unigenes) in our dataset by assembling only the reverse reads of the directionally cloned cDNAs. The resulting estimate of 2,564 unigenes compared to the 4,925 UniSeqs is likely to reflect the large average size (1.95 kb) of inserts in the cDNA library; thus, in many cases, UniSeqs represent the 3' and 5' ends of genes for which the central parts were not captured due to Sanger-sequencing length limitations (600–800 bp). In addition, different splice variants or alleles of the same gene may have contributed to the excess of UniSeqs over unigenes. Detailed pre-assembly statistics are summarized in Additional file 1: Quality control and assembly statistics.

Previously, a small-scale EST project was conducted in order to compare the abundance of transcripts between symbiotic and aposymbiotic Aiptasia pulchella polyps 35. The present study included bi-directional sequencing, and the total number of ESTs is more than 14 times larger than in the earlier study. Therefore, the availability of almost 5,000 UniSeqs for about 2,500 unigenes represents a rich transcriptomic resource, previously unavailable at this scale, for a symbiotic anemone.

All UniSeqs were assigned putative identities based on BLASTx hits (E-value cutoff: 1e-5) to the UniProt Knowledgebase databases SwissProt and TrEMBL 41. About 37% and ~63% of the UniSeqs found hits in SwissProt and TrEMBL, respectively, leaving ~36% of the UniSeqs without similarities to known proteins. Because the TrEMBL database contains protein sequences based on conceptual translations of all nucleotide sequence entries in EMBL/GenBank/DDBJ, we chose to annotate the UniSeqs according to the curated SwissProt entries. Assignments of gene ontologies (GO) could be made for about one third of UniSeqs in each of the GO categories: molecular function, biological process, and cellular component. Because our cDNA library represents the symbiotic, adult life-history stage of A. pallida, the GO resource generated in this study sets the stage for statistical assessments of over- or under-representation of specific GO-categories in libraries obtained from anemones under different conditions such as life-history stages, symbiotic state (symbiotic vs. aposymbiotic), or environmental conditions (temperature, salinity, nutrients, etc.). In addition to BLAST and GO annotations, all UniSeqs were screened for single-nucleotide polymorphisms (SNPs) and simple-sequence repeats (SSRs), providing resources for the investigation of gene polymorphisms between individuals and/or populations. The prediction of open reading frames within UniSeqs also provided the basis for domain annotations at the protein level. About 25% of UniSeqs matched a protein domain entry in the Pfam database 42.

One of the primary challenges of sequencing ESTs from a mixed transcriptome originating from two or more partners is to assign sequences to the proper genome of origin. Taking a bioinformatic approach to this problem, we constructed taxon-specific databases representing either "Cnidaria-only" or "Alveolata-only" (i.e., dinoflagellates and their relatives) sequences from GenBank, and then performed BLASTx-searches against those databases as well as the complete non-redundant database (see Methods). We then employed a best-BLASTx-hit (BBH) approach (Additional file 2: Flow diagram illustrating BBH approach) to estimate the numbers of ESTs that originated from A. pallida and Symbiodinium spp., respectively, at various levels of confidence (Table 1). Irrespective of the confidence level, about one quarter of ESTs had no BLASTx-hit (E-value cutoff 1e-5). At the different levels of confidence, 56 – 70% and 1.7 – 6.4% were predicted to originate from the anemone and the Symbiodinium genomes, respectively (Additional file 3: Detailed EST (N = 10,285) distribution and assignment). The relatively small fraction of Symbiodinium ESTs could be expected given that Symbiodinum spp. are spatially restricted to the endodermal tissue layer of the host and that no special effort was made to disrupt the algal cell walls during the preparation of the RNA (see Methods). Furthermore, the number of UniSeqs without a significant BLASTx hit may be higher for Symbiodinium transcripts. However, the uncertainty about the origin of non-annotated sequences represents a current limitation to our approach. Ongoing and future genome-sequencing projects for symbiotic cnidarians and their dinoflagellate endosymbionts should soon become available and help to uncover the origins of sequences without currently known homologs in other organisms. This will provide an interesting opportunity to revisit our data set to look further at these perhaps taxonomically restricted genes.

Using an EST-processing software (EST2uni) 36, we stored all ESTs, UniSeqs, and annotations in a queryable database named AiptasiaBase (database = AiptasiaBase_v1), which is accessible through the URL: http://aiptasia.cs.vassar.edu/AiptasiaBase/index.php. In addition to the results generated by the software, we have included the annotation of UniSeqs according to KEGG, which provides a convenient way to explore pathway components that were identified in this study.

We identified the contigs containing the greatest numbers of ESTs, which we used as a proxy for the most abundant transcripts. Although the numbers of ESTs in contigs that are predicted to originate from Symbiodinium were too low to be analyzed (data not shown), many of the most abundant host-derived transcripts represented genes that are involved in the processes of protein biosynthesis, extracellular-matrix formation, and oxidative-stress response (Table 2).

Kuo et al. (2004) reported that the most highly expressed gene in symbiotic A. pulchella was ferritin (11.7%), whereas we found only 4 ESTs (0.04%) that represented this gene. Although differences in the preparations of the cDNA libraries (e.g. insert size-selection) and sequencing depths (474 vs. 10,285 ESTs) pose an obstacle for a direct comparison, the discrepancy in the numbers of ferritin transcripts appears to be noteworthy. In a recent study that investigated the effect of increased temperature and UV levels on the symbiotic anemone Anthopleura elegantissima, Richier et al. (2008) observed a more than 17-fold up-regulation of ferritin expression upon thermal stress, but not UV stress 43. Given this observation, it seems possible that the anemones in the study by Kuo et al. (2004) were under elevated thermal stress at the time of sampling, which, taken together with the methodological differences mentioned above, makes any further comparative analyses unfeasible. This result has important implications, i.e., how culturing conditions of organisms as well as methodological differences between studies may have an impact on the transcriptome, and by extension, the interpretation of gene expression analyses.

The highly abundant sequences with the highest uncertainties for correct annotation (highest E-values), apolipophorin and the CUB and zona pellucida-like domain-containing protein 1, were further scrutinized by similarity searches in additional databases. These searches revealed that the best hit for CUB and zona pellucida-like domain-containing protein 1 in the GenBank non-redundant database (nr) was mesoglein, a protein that is proposed to be a structural element of the extracellular matrix of the mesoglea in the jellyfish Aurelia aurita 44. The sequence annotated as apolipophorin contains a von Willebrand factor type-D domain, and was reported to be involved in forming lipoprotein particles that bind lipoproteins and lipids 45. Two other highly abundant sequences had no homologs among previously characterized proteins, suggesting that they are novel, and a third contig with no BLASTx-hit was identified as an artifact due to misassembled sequences. These results illustrate some of the caveats to automated sequence assembly and annotation and highlight the necessity for corroboration after automated sequence processing when focusing on single genes or groups of genes of interest.

We generated a candidate gene list of groups containing UniSeqs that are likely to be of relevance to cnidarian-dinoflagellate symbioses (Table 3). Among these, the cellular antioxidant-response system could be most comprehensively reconstructed (see below). Genes related to the innate immune system and sugar-binding proteins gave rise to a partial gene inventory (Fig. 1; Table 3). Other genes that are likely to play a role in the cellular events surrounding the breakdown of symbiosis (exocytosis, host-cell detachment, apoptosis and/or autophagy 46474849505152) were also identified.

Stress-induced photoinhibition and damage to algal photosystem II are thought to be responsible for an increased production of reactive oxygen species 5354 and consequently, diffusion of hydrogen peroxide (H₂O₂) through the membranes into the host cells 55. The detoxification of H₂O₂ requires the activity of catalase or other peroxidases. Superoxide dismutase (SOD), which catalyzes the reduction of superoxide to H₂O₂, as well as glutathione peroxidase (GPx), which uses glutathione to detoxify H₂O₂, were both not found among the sequenced ESTs. One possibility is that the abundance of SOD transcripts in host cells was low, and the generation of superoxide spatially limited (inside the chloroplasts of Symbiodinium). In this case, superoxide may have been efficiently eliminated within the Symbiodinium cells, while excess H₂O₂ that was not detoxified (e.g., by Symbiodinium ascorbate peroxidase), could have diffused into the host cytosol and been reduced to H₂O and O₂ by catalase. Alternatively, methodological factors such as insert-size selection or general RNA processing may have prevented the detection of SOD. Other genes that had previously been reported in the context of cnidarian-dinoflagellate symbioses (Additional file 4: Genes that have been studied in the context of cnidarian-dinoflagellate symbiosis, but not found in this study) were also not detected, perhaps for same reasons as discussed above for SOD.

A clonal line of Aiptasia pallida (clone CC7, available through the Pringle lab) hosting Symbiodinium of clade A was established from a single tiny propagule in a population obtained from Carolina Biological Supply (Burlington, NC) and grown into an abundant stock. Given the Symbiodinium clade harbored by this population, it is likely that the Aiptasia individual originated from the Florida Keys lineage. Approximately 500 anemones of various sizes were harvested from this stock under normal growth conditions (~26°C; salinity, ~33 ppt; light, ~40 μmol m^-2 s^-1 photosynthetic photon flux; 12-h light-dark cycle), blotted to remove excess water, and immediately frozen in liquid nitrogen. The anemones were then ground to a fine powder under liquid nitrogen using a ceramic mortar and pestle. The powder was weighed (~4 g) while still frozen and mixed with a proportional volume (50 ml) of TRIzol Reagent (Invitrogen, Carlsbad, CA); extraction was then performed in accordance with the manufacturer's instructions yielding ~5 mg of total RNA. This RNA was sent to Open Biosystems (Huntsville, AL), where it was tested for quality; mRNA was then isolated using oligo(dT)-coated magnetic particles (Seradyn, Indianapolis, IN), and cDNA was synthesized. Double-stranded cDNA was size fractionated to enrich for long reads, cloned into the vector pExpress1 (Express Genomics, Frederick, MD), and electroporated into E. coli strain DH10B. The resulting library was determined to contain ~96% recombinants with an average insert size of 1.95 kb. Sequencing was performed on 96-well capillary sequencing platforms (ABI 3700) at the DOE Joint Genome Institute (JGI, Walnut Creek, CA) and at the Genome Core Facility at the University of California, Merced, USA, CA.

Raw chromatogram files were used as input for the software pipeline EST2uni 36, which was implemented on an Ubuntu server (8.04 "Hardy Heron", Dual Intel Xeon 3.06 GHz) to generate the database named AiptasiaBase 57. During the pipeline processing, raw EST reads were based-called by phred 37, and quality filtered and vector trimmed by the software Lucy 38; low-complexity regions and repetitive elements were then removed by seqclean 39 and RepeatMasker 58, respectively. To remove unexpected vector sequences, seqclean additionally screened the processed ESTs using NCBI's UniVec database. All ESTs are available through GenBank accession numbers GH571982 – GH582266.

Clustering of processed ESTs was performed by cap3 40 with default settings resulting in unique sequences (UniSeqs), for which open reading frames were predicted by ESTScan 59. Similar UniSeqs were found using BLASTn 60, resulting in clusters of similar UniSeqs 60. Short-sequence-repeat microsatellites and sequence variations were predicted by Sputnik 61 and local algorithms 36, respectively. All UniSeqs were functionally annotated by BLASTx searches 60 in protein databases nr (GenBank – NCBI), TrEMBL, and SwissProt (Uniprot) 62; HMMER 63 searches in pfam 42; and GO-term associations (UniProt GOA, March 2008) 64. The number of unique genes was estimated by clustering all reverse reads using the cap3 software with default settings.

In order to predict whether an EST originated from Aiptasia pallida or Symbiodinium spp., we performed a best-BLASTx-hit (BBH) approach (Additional file 2: Flow diagram illustrating BBH approach). First, all UniSeqs were BLASTx-searched (E-value cutoff: 1e-5) in a non-redundant protein database (nr, GenBank, NCBI). If the BBH was from a cnidarian or an alveolate species, the sequence was predicted to originate from Aiptasia pallida or Symbiodinium spp., respectively, with high confidence. Next, if the BBH was not from a cnidarian or alveolate species, we compared the E-values for the BBHs from a search against nr databases that were previously filtered for sequences from cnidarian (582,480) or alveolate (468,072) species. The organism for which the E-value was lower was assigned to the corresponding UniSeq with medium confidence. Finally, if the E-values for BBH searches in the cnidarian and alveolate databases were equal, we compared the percentage of identical amino acids in the sequence alignments. As in the E-value-based approach, the organism with the higher percentage of identical amino acids was assigned to the corresponding UniSeq (low confidence). In addition to the annotations described above, we used the Automatic Annotation Server provided by the Kyoto Encyclopedia of Genes and Genomes (KEGG) for all UniSeqs using the single-directional best-hit option.