Transcriptome sequencing offers the possibility to discover novel expression patterns, high abundance transcripts, sequence motifs and splice variants 6. In addition transcriptome sequencing measures abundance at a wider scale, as this technique does not get oversaturated. However challenges regarding quality do exist, it is for example crucial for the correct annotation to have a high-quality reference genome. Another aspect of quality is how reliable the expression patterns discovered in libraries really are. Highly expressed genes represent processes that dominate in the given developmental stage, and discovery and measurements of such gene expressions are more likely to be accurate than that of genes that are weakly expressed. Singletons (n = 1) are more likely to represent random sequencing or sequnecing errors than transcripts that are sequenced numerous of times for one gene. Likewise, in microarray studies, low number of transcripts causes a large variation in gene expression results 23. This was also confirmed in this study as we revealed that for genes with few sequences (total n = 1-5) detected in the libraries, the expression profile from oocyte until gastrula could not be reflected in expression profiles measured with quantitative PCR (qPCR) for the same genes (Figure 2). For genes with a higher number (n = 12-61) of sequences, their expression profiles corresponded with the ones measured by qPCR (Figure 2). We therefore suggest for further transcriptomic studies to filter data based on analysis of number of transcripts to obtain reliable expression profiles for downstream analyses. A threshold of 9 or more total sequences for each gene was selected to be applied for our dataset, to be able to include only the most reliable data for analysis. Threshold 9 was chosen because it is in the middle of 5 and 12, and using this threshold did not considerably affect number of genes to be excluded for downstream analyses compared to when applying threshold 5. Given that only 1 EST was sequenced for approximately 50% of the genes in our dataset (data not shown), even a very low cutoff would exclude a high number of genes from the analysis. However some important information may be hidden within genes with low expression, and will therefore be lost by applying an EST number threshold. For example, the zebrafish oocyte possesses a considerable richness of maternally expressed genes of which many may be of importance for embryogenesis 6. When the current experiment was carried out the cod genome was not yet sequenced. However, using the current genome 24 together with new sequencing technology could significantly increase number of reads analyzed in this study. For example, Aanes recently studied the zebrafish embryonic transcriptome by applying RNA sequencing 6, a method which provides a considerably more precise measurement of transcript abundance due to high number of sequence reads 202122. If we were to perform this study again with RNA sequencing, more sequence reads would be obtained and more genes could be included in the analysis as they would be represented by more than only 1 or a few sequenced transcripts.

Also if candidate gene expressions from cDNA libraries are to be verified by qPCR, stable reference genes are essential. When it comes to early embryo tissue like in this study, such stable references are difficult to obtain. For example, neither β-actinelongation factor 1- α (ef1α)18S, actin related protein (arp) nor ubiquitin (ubi) showed stable reference gene expression in stages assayed for gene expression [Additional file 1, 2526. For this reason we chose not to use any of the reference genes tested, but instead apply a technique (external spiking) where external RNA (rabbit alpha hemoglobin (hba)) is added in equal amounts to each RNA sample 27. The considerably more stable expression pattern of hba can be seen in [Additional file 1. We therefore suggest that external spiking may serve as a good alternative to the traditional reference genes previously applied in qPCR, especially when measuring gene expressions in early embryo tissue.

In order to get an impression of the approximate timing of MZT (and therefore the shift from degrading to activating transcripts) we measured gene expression profiles of candidate genes known from other species to be maternal, zygotic or involved in events taking place at the initiation of MZT [Additional file 2. Our results together with the findings of Drivenes 19 indicate that the MZT takes place between blastula and gastrula in Atlantic cod, which is also true for zebrafish 12. Together the cDNA libraries from oocyte, 1-2 cell, blastula and gastrula stage produced 26162 3’EST sequences, which have been added to GenBank; 6511, 6607, 6506 and 6538 sequences were obtained from oocyte, 1-2 cell, blastula and gastrula libraries, respectively. 22229 sequences (5538, 5875, 5238 and 5578 in oocyte, 1-2 cell, blastula and gastrula, respectively) were mapped to the Atlantic cod genome 24. 17025 of these sequences hit cod genes with the rules given in material and methods section; they were predicted to encode 2611 genes, of which 2051 (78.6%) were Ensembl-annotated. 23 genes were pseudogenes, and were therefore removed from most of the analysis. Furthermore as described in methodological considerations, sequences with n ≤ 8 were removed from most of the analysis, leaving us with 12764 sequences predicted to be encoded by 257 genes. Compared to similar analyses 202122, 257 genes is a low number; however these studies have included most transcripts in contrast to the current study. As mentioned, we showed no correlation between qPCR and low copy number of ESTs (N < 9, see Figure 2 and discussion first section; methodological considerations), and highly expressed genes (N > 9) are possible to classify as maternal, zygotic, degrading and activating. EST copy number filtering was a necessity to be able to study stability and complexity of dominating transcripts in early embryos of cod.

Strikingly, the overall composition of abundant transcripts obtained from maternal stages (oocyte, 1-2 cell and blastula) contained 45.9-50.6% mitochondrial transcripts, while at the zygotic stage (gastrula) only 12.2% of the transcripts were mitochondrial (Figure 3). Less than 5.1% of all the expressed genes in each developmental stage encoded these numerous mitochondrial transcripts (data not shown). This trend was also evident by the highly expressed mitochondrial genes of Table 1, showing the 20 most abundantly expressed genes when all transcripts from all libraries were included. Likewise it has been reported a high level of mitochondrial transcripts at the 2-cell stage in halibut 22 and in the oocyte of Senegalese sole 28. High level of mitochondrial transcripts in mature oocytes may be linked to high metabolic need to complete synchronous continuous cell-divisions at early developmental stages, whereas at later stages the energy metabolism is more localized to target areas in specific development. It has even been suggested that a major part of maternal RNAs in egg-laying organisms lack specificity and only provides supplemental nutrition 29. However a proportion of the maternal RNAs are essential for normal development in fish, as several studies report abnormal embryo development when mRNA translation is inhibited 30. Also several processes in zebrafish morphogenesis, most notably brain formation, fail to occur normally when normal degradation of maternal RNA is disturbed 31.

Gene names and their total number of transcripts in oocyte, 1-2 cell, blastula and gastrula are shown.

To facilitate analysis of expressed genes from the genome, mitochondrial transcripts (4786, encoded by 10 genes) were excluded. What remained were the nuclear transcripts (7978, encoded by 247 genes). A significant feature of the gastrula nuclear transcriptome was a high presence of ribosomal transcripts (50.6%) compared to the earlier stages (2.5–4.6%, Figure 4a). This increase in transcript abundance was not a consequence from many additional genes being expressed in the gastrula stage, as the number of genes encoding the ribosomal transcripts only increased from 6.7–13.1% in the maternal libraries to 19.7% in the gastrula library (Figure 4b). However, some of the ribosomal transcripts were uniquely found in the gastrula library (Table 2). A similar pattern of ribosomal transcripts has previously been found in zebrafish 17 and halibut 22. The high presence of ribosomal transcripts in gastrula is most likely due to major onset of transcription since ribosomal proteins are required for the production of ribosomes at the site for transcription.

Gene names and their total number of transcripts in oocyte, 1-2 cell, blastula and gastrula are shown.

Most nuclear genes (79.4%) were expressed in more than one developmental stage, and 51.8% (128) of the genes were expressed in all four stages. The most highly expressed genes (see Table 1) present in all libraries were histone H2A family member V followed by tetraspanin 3, which are involved in general processes like structuring of chromosomal fibres and signal transduction in cell development, activation, growth and motility. 12.1% (30) of the nuclear genes were expressed exclusively in maternal stages (oocyte only or oocyte and one or two of the next developmental stages), and represented a diverse group (Table 3). However among the ones with the highest numbers of transcripts in this category, we found common genes (like COX17stathmin and cyclins) involved in processes like energy metabolism and cell cycle, which are features that characterize early embryonic development. Other studies 62232 report on a higher number of strictly maternal genes than 30, however as mentioned many genes were excluded in this study due to their low number of sequenced ESTs. Since the cutoff of N = 9 may have excluded many lower expressed but important maternal genes from downstream analysis all sequenced genes were added to GenBank, leaving them available for future studies of for example egg viability.

Gene names and their total number of transcripts in oocyte, 1-2 cell, blastula and gastrula are shown.

We also checked how the nuclear transcripts were distributed when including all nuclear genes (2575) without any threshold for number of transcripts. In this dataset a large proportion (56.5%) of the genes had only 1 transcript. Consequently, the genes expressed exclusively in maternal and gastrula stages showed an increase in their proportions (12.1–31.9% and 6.9–13%, respectively). The proportion of genes expressed in all developmental stages decreased markedly from 51.8 to 6.5% [Additional file 3]. It is therefore clear that the use of a threshold for number of transcripts makes a difference for describing gene expression patterns.

Functional profiles of the most common nuclear transcripts in oocyte, 1-2 cell, blastula and gastrula (obtained from the use of gene ontology (GO)-terms) revealed that both within cellular component, molecular function and biological process many of the transcripts did not change much in their abundance over time (data not shown). However some differences could be observed between maternal and zygotic stages [Additional file 4. An increase in proportion of transcripts that were part of macromolecular complexes from maternal to zygotic developmental stage possibly reflects specification of cell types which requires more complex molecules as the embryo enters the zygotic stage. At the molecular function level, a pronounced increase of transcript amount from maternal to zygotic stage was evident for structural molecule activity, reflecting a shift towards specification of cell function. Several biological processes including cellular component organization or biogenesis, multicellular organismal process and developmental process showed a marked shift between maternal and zygotic stage, with increase in transcript proportion in gastrula. Taken together, these changes reflect more specialized and complex processes developing after MZT in cod, as would be expected since gastrula is a developmental stage where intricate processes like germ layer formation takes place (reviewed in 33).

Based on manual annotation, 3’UTR length was predicted for all sequences with stop codon and poly-A tail (in total 191 genes). To be able to study the gene expression dynamics over time from a maternal to a zygotic embryo, genes (when possible) were strictly classified as having degrading, stable or activating transcripts. In order to be termed degrading, minimum 95% of the transcripts had to be located in maternal stages (oocyte until blastula). In order to be termed activating, minimum 95% of the transcripts had to be present in blastula and gastrula. Transcripts present in all 4 stages, with a distribution of minimum 10% and maximum 40% in one stage were termed stable. Applying this classification gave 35, 26 and 39 genes with degrading, stable and activating expression profiles, respectively. 91 genes could not be classified according to these specifications. To ensure reliable results, the following analyses on 3’UTR structure were also performed when applying less strict specifications [Additional file 5].

Totally 73 nuclear genes (29.6%) expressed more than one 3’UTR isoform. Among genes with degrading and stable transcripts a relatively high amount (45.7 and 34.6%, respecively) displayed 3’UTR isoforms (Figure 5). Alternatively spliced transcripts are common in early embryos in other species, for example in zebrafish where it has been shown 50–60% alternative splicing of transcripts 6. Also in mouse embryos a significant proportion of alternatively spliced transcripts have been found 34, and in this context it was suggested that alternative splicing is of importance in the regulation of development. Activating genes of cod had a far less degree of 3’UTR isoforms (15.4%, Figure 5). This may be an indication that alternative splicing is a less important regulatory mechanism for genes that are more active in early zygotic stages, than for genes that are more active in maternal stages.

Mean 3’UTR length (when including all isoforms) in degrading (n = 48), stable (n = 35) and activating (n = 45) transcripts was 187.1, 208.8 and 76 bp, respectively (Figure 6). Activating transcripts had significantly (p < 0.0001) shorter 3’UTR length than degrading and stable ones. A higher proportion of long maternal 3’UTRs have previously been found across the animal kingdom 29 associated with lower turnover-rate of RNAs 35. Our findings suggest that also in cod, transcripts from maternally active genes have longer 3’ UTRs. The shorter 3’ UTRs of the activating transcripts may also be explained by the fact that many of them were ribosomal (data not shown); such transcripts have been shown to contain short 3’UTRs 36. Excluding ribosomal genes would lead to a small dataset. Consequently, we applied less strict classification criteria for degrading, stable and activating transcripts and found that also non-ribosomal activating transcripts have significantly shorter 3’UTRs [Additional file 5.

Although not statistically proved in this study, stable transcripts appeared to have the longest 3’UTRs (Figure 6). This corresponds to the suggestion that more stable RNAs have longer UTRs 37. A strong exponential correlation has been found between 3’UTR length and morphological complexity (number of cell types in the organism) 38. Increased regulatory complexity in long 3’UTRs may be a result of more miRNA binding sites. Furthermore, additional polyadenylation signals may produce transcript isoforms with several combinations of miRNA binding sites, which adds even more complexity 39. It was observed an enrichment of 3’UTR isoforms in degrading and stable transcripts (abundant in the maternal stages) compared to activating ones (abundant in zygotic stage). This result suggests a more complex post-transcriptional regulation in the maternal stages since transcriptional regulation cannot be applied to control developmental processes prior to MZT.

The apparent complex features of transcripts that were abundant in early developmental stages of cod were further investigated by searching for motifs in the 3’UTRs of degrading, stable and activating transcripts. Using Multiple Em for Motif Elicitation (MEME) 40 the presence of the polyadenylation motif AATAAA was indicated. Manual inspection of the sequences revealed that AATAAA (the exact motif) was present at an increasing manner from degrading (68.8%), through stable (82.9%) and to activating (95.6%) transcripts (Figure 7a). AATAAA is essential for cytoplasmic polyadenylation to take place 4142; that is, cleavage followed by adding of a poly-A tail to the 3’ end of the mRNA which then activates translation. Therefore polyadenylation is a key process in regulation of translation (for recent reviews see 4344), and thus serves as an important step in post-transcriptional regulation in maternal stage embryos. Also, it was recently found that cytoplasmic polyadenylation of maternal RNAs prior to MBT is essential for normal morphogenesis post-MBT in zebrafish 6. It is known that AATAAA is found in nearly all known mRNAs, and that it is highly conserved. Our finding that 31.2 and 17.1% of degrading and stable transcripts (with known 3’UTR length) did not contain AATAAA may be explained by the possibility of transcripts to contain alternative polyadenylation motifs, where AATAAA have been slightly modified (reviewed in 43). Therefore, we investigated the presence of 11 other putative alternative polyadenylation motifs 45; for each sequence we observed which motifs were present, and if there was more than one motif (Figure 7). 3 motifs (ATTAAA, AATATA and AATAGA) were found exclusively in degrading and stable transcripts, and AGTAAA was present only in stable transcripts. None of the motifs were unique for activating transcripts, and most of the motifs were present at the lowest rate in this group of transcripts (Figure 7a). Also, degrading and especially stable transcripts had a higher prevalence (41.7 and 54.3%, respectively) of multiple motifs (more than one of the motifs studied) than activating ones (31.1%, Figure 7b). A higher degree (diversity and quantity) of putative alternative polyadenylation motifs in maternal degrading and stable transcripts suggests that these transcripts may apply alternative polyadenylation more frequently than activating transcripts. Polyadenylation at different sites produces different variants of the 3’UTR, which contributes to a more complex transcriptome 46.

Eggs and sperm from Atlantic cod were obtained from Parisvatnet Research station, Institute of Marine Research, Øygarden, Norway in April 2009. Eggs were manually stripped from one female and sperm was collected from one male. Before freezing of unfertilized eggs, egg fluid was carefully aspirated and subsequently eggs were flash frozen liquid nitrogen. Remaining eggs were fertilized with the collected sperm and incubated at 7°C until they reached 1-2 cells, blastula and gastrula stages (Figure 1), embryos were staged according to Hall 47. At sampling, staged embryos were collected and excess water was aspirated off before embryos were flash-frozen on liquid nitrogen. All collected samples were stored at −80°C until RNA isolation was initiated. All egg and embryo samplings were done according to guidelines approved by the Norwegian Animal Research Authority (NARA).

From eggs and each developmental stage, total RNA was isolated using iPrep™ Trizol® Plus RNA Kit (Invitrogen, Carlsbad, CA, USA). The quality of RNA showed RIN values above 9.5 for all samples used (Bioanalyzer 2100, Agilent, Santa Clara CA, United States). cDNA libraries were constructed from 1.0 μg of RNA from each sample using Creator SMART cDNA Library Construction Kit (Clontech Laboratories, Inc., Mountain View, CA, USA). The cDNA was size-selected for 500bp according to Evsikov 21. Transformation of the ligation mix was performed using the DH5αMAX competent cells (Invitrogen, Carlsbad, CA, USA). The titers of libraries used were for oocytes 7,7x10⁵, 1-2 cell stage 4,7x10⁵, blastula 1,5x10⁶ and gastrula 1,4x10⁶ clones per μl of ligation mix. The clones were grown on robot plates (Nalgene Nunc International, Rochester, NY, USA) filled with Luria-Bertani (LB) agar and chloramphenicol (30mg/L). Positive clones were randomly selected by automated picking robots and transfered to 384-well plates containing HMFM freezing media. Culturing was performed for 16h at 37°C without agitation. For isolation and purification of plasmid DNA 384-deepwell plates containing 190μl 2YT media and 30 mg/L chloramphenicol were inoculated from the freezing stocks. Clones were grown for 16-18h at 1100 rpm and 37°C. Isolation and purification of plasmids was done by an automated system, applying alkaline lysis followed by size selective DNA precipitation in PEG/2-Propanol 48. Quantified plasmids (UV photometry) were sequenced using T7 primers following the Big-Dye version 3.1 protocol and capillary gelelectrophoresis by an ABI 3730xl system (Applied Biosystems Inc., Foster City, Ca, USA) at Max Planck Institute for Molecular Genetics, Berlin, Germany.

Sequence trace files from each library were processed employing the Phred base-calling algorithm 4950. Quality trimming and removal of contaminating vector sequences were performed using cross match and RBR 51 with the Uinivec filtering database from the National Center for Biotechnology Information (NCBI). Clustering and assembly were carried out using the TIGR Gene Indices Clustering tools (TGICL) 52. Sequences for the libraries were added to GenBank; oocyte library [GenBank: JK777860-JK784370], 1-2 cell library [GenBank: JK771253-JK777859], blastula library [GenBank: JK758209-JK764714] and gastrula library [GenBank: JK764715-JK771252], respectively.

Clustered sequences were assembled into contigs (Fasta sequences are given in [Additional file 6) which were further annotated with Ensembl predicted genes 53 by mapping the contigs to the Atlantic cod genome 24 with a threshold above 75% identity BLAT hit against a genome scaffold in a gene or within 2000 bp downstream of a gene. Sequences that passed the following rules were included in further analysis: 1) above 85% identity BLAT hit against a genome mapped EST-contig, or 2) above 75% identity BLAT hit against a genome scaffold in a gene or within 2000 bp downstream of a gene. The mapping of the contigs can be seen in [Additional file 7.

Functional profiles of transcripts were constructed by applying GO-terms to group the transcripts 54. The Ensembl genes were GO-annotated using BLAST against SwissProt and UniRef90 and annotation files from 55. BLAST and GO files for SwissProt and UniRef90 that are constructed for Ensembl-annotated genome assembly can be seen in [Additional file 8 and [Additional file 9, respectively.

Total RNA was extracted by applying RNeasy® Mini Kit (QIAGEN, Oslo, Norway), and genomic DNA was removed by using Turbo DNA-free^TM Kit (Ambion, Austin, Texas, USA), according to the manufacturers instruction. Additionally, all samples were precipitated and redisolved in ddH20 after DNAse treatment. A NanoDrop® NP-1000 spectrophotometer (NanoDrop technologies, Wilmington, DE, USA) was used to measure the quantity and quality of the RNA samples. Samples with a 260/280 nm absorbance ratio outside the range 1.8–2.1 were excluded for further analysis. 1 pg of rabbit hba mRNA (SIGMA; Norway) was added to each RNA sample before cDNA synthesis (spiking). cDNA was produced using Superscript VILO cDNA synthesis (Invitrogen, Carlsbad, Germany). All primers used for amplification and detection of genes were designed applying the software Primer Express 3.0 (Applied Biosystems, Foster city, CA, USA), and are listed in Table 4 together with hba-primers 27. Quantitative PCR (qPCR) was performed on an SDS 7900HT Fast Real-Time PCR system (Applied Biosystems, Oslo, Norway) system using SYBR® green PCR Master Mix (Applied Biosystems, Foster city, CA, USA), and the thermal cycling conditions were: 50°C for 2 min followed by 98°C for 10 min, and 40 cycles of 95°C for 15s followed by 60°C for 1 min. PCR efficiencies were verified to be equal between target and calibrator genes (Standard-curve method using 250, 125, 62.5, 31.25 and 15.625 ng RNA). Melting curve analysis was performed to verify presence of only one PCR product. 62.5 ng RNA was used to produce cDNA used for downstream analysis of relative abundance of transcript. For each PCR plate, no-template controls were run for each gene. The relative gene expression level was calculated using the Comparative Ct method (Applied Biosystems). All data are normalized to hba mRNA (spike), and the data for each gene are calibrated to its lowest mean ΔCt.

When possible, sequences are listed with an Ensembl gene ID. Otherwise the sequences are listed with GenBank EST ID (orp and siva) or mRNA ID (hba).

All data to be analyzed for significant differences were initially tested for Gaussian distribution by applying a D’Agostino-Pearson normality (omnibus K2) test. All mRNA levels quantified by qPCR had too few replicates (n = 3) for each developmental stage in order for the test to reveal normality or not. Therefore a non-parametric Kruskal-Wallis test was applied to reveal if the genes were differentially expressed in one or more of the developmental stages. Dunn’s multiple comparison post-test was used to detect at which developmental stages there were differences. Furthermore there was no evidence for Gaussian distribution in the datasets for 3’ UTR length analysis, including when they were converted to log values. Therefore a non-parametric Kruskal-Wallis test was also applied here to calculate if there were any differences in UTR length between different groups of transcripts. Dunn’s post test was applied to reveal which groups differed from each other. The statistical tests were performed using GraphPad Prism 5.04 (GraphPad Software Inc., La Jolla, CA92037, USA). A p-value of ≤0.05 indicates significant difference.