Detection of RNA structures in porcine EST data and related mammals
- 6.4k Downloads
Non-coding RNAs (ncRNAs) are involved in a wide spectrum of regulatory functions. Within recent years, there have been increasing reports of observed polyadenylated ncRNAs and mRNA like ncRNAs in eukaryotes. To investigate this further, we examined the large data set in the Sino-Danish PigEST resource http://pigest.ku.dk which also contains expression information distributed on 97 non-normalized cDNA libraries.
We constructed a pipeline, EST2ncRNA, to search for known and novel ncRNAs. The pipeline utilises sequence similarity to ncRNA databases (blast), structure similarity to Rfam (RaveNnA) as well as multiple alignments to predict conserved novel putative RNA structures (RNAz). EST2ncRNA was fed with 48,000 contigs and 73,000 singletons available from the PigEST resource. Using the pipeline we identified known RNA structures in 137 contigs and single reads (conreads), and predicted high confidence RNA structures in non-protein coding regions of additional 1,262 conreads. Of these, structures in 270 conreads overlap with existing predictions in human. To sum up, the PigEST resource comprises trans-acting elements (ncRNAs) in 715 contigs and 340 singletons as well as cis-acting elements (inside UTRs) in 311 contigs and 51 singletons, of which 18 conreads contain both predictions of trans- and cis-acting elements. The predicted RNAz candidates were compared with the PigEST expression information and we identify 114 contigs with an RNAz prediction and expression in at least ten of the non-normalised cDNA libraries. We conclude that the contigs with RNAz and known predictions are in general expressed at a much lower level than protein coding transcripts. In addition, we also observe that our ncRNA candidates constitute about one to two percent of the genes expressed in the cDNA libraries. Intriguingly, the cDNA libraries from developmental (brain) tissues contain the highest amount of ncRNA candidates, about two percent. These observations are related to existing knowledge and hypotheses about the role of ncRNAs in higher organisms. Furthermore, about 80% porcine coding transcripts (of 18,600 identified) as well as less than one-third ORF-free transcripts are conserved at least in the closely related bovine genome. Approximately one percent of the coding and 10% of the remaining matches are unique between the PigEST data and cow genome. Based on the pig-cow alignments, we searched for similarities to 16 other organisms by UCSC available alignments, which resulted in a 87% coverage by the human genome for instance.
Besides recovering several of the already annotated functional RNA structures, we predicted a large number of high confidence conserved secondary structures in polyadenylated porcine transcripts. Our observations of relatively low expression levels of predicted ncRNA candidates together with the observations of higher relative amount in cDNA libraries from developmental stages are in agreement with the current paradigm of ncRNA roles in higher organisms and supports the idea of polyadenylated ncRNAs.
KeywordsOpen Reading Frame Region ncRNA Family ncRNA Candidate Putative ncRNAs Reverse Complementary Strand
Genomic studies show that only a small proportion of transcribed RNAs represent messenger RNAs . Less than 2% of the human genome codes for proteins, although a large fraction of the eukaryotic genome is transcribed [2, 3, 4, 5]. In fact, the ENCODE Pilot Project provides evidence that almost the entire non-repetitive part of the human genome is transcribed in at least one of the two reading directions . In agreement with this, in recent years, there has been reported an increasing number of functional non-coding RNAs (ncRNAs) . The discrimination between functional ncRNAs and genomic transcription background is a complex problem  since ncRNAs do not present common primary sequence features. Furthermore, recent results show that very short ORFs may be translated to yield functional proteins , emphasizing that the absence of a long open reading frame alone does not necessarily imply that a transcript functions as ncRNA.
Most of the "house keeping" ncRNA families (tRNAs, rRNAs, snRNAs, snoRNAs) and a large class of regulatory RNAs (in particular miRNAs) have characteristic structures which perform an evolutionary conserved function. This property can be utilized in comparative genomics approaches to recognize such functional RNAs [10, 11, 12, 13]. Such computational surveys have resulted in the prediction of many thousands of genomic loci with evidence for stabilizing selection of RNA structures [14, 15, 16, 17, 18, 19, 20].
Despite the relatively high false positive rates of ncRNA predictions by programs such as EvoFold, RNAz, and foldalign all these approaches yield clear statistical evidence that evolutionary conserved RNA structure is a pervasive feature of eukaryotic genomes. The experimental verification of the predictions is a complex issue. Northern blots, the method of choice for this task, require thousands of copies of the RNA molecule for a detectable signal. Non-coding transcripts, however, appear to be expressed at much lower levels than most protein-coding mRNAs, see e.g. . Furthermore, some ncRNAs might only be expressed for a short time at a particular developmental stage or only in a very specific tissue. An extreme example is the microRNA lsy-6 in C. elegans, which is expressed only in a single neuron of the worm .
Expressed sequence tags (ESTs) represent short subsequences of transcribed RNAs. They are produced by an automated and cost effective sequencing mechanism that, however, results in low quality sequences which typically do not cover the complete transcript. The majority of EST data comprise poly(A)(+) RNAs since the cDNAs are obtained using a poly-T primer. Polyadenylation is often thought of as a characteristic feature of messenger RNAs (mRNAs). Mammalian transcriptomes, however, contain thousands of mRNA-like RNAs that are spliced but do not have appreciable ORFs or other evidence for protein coding capacity. This class of transcripts constitutes a significant fraction of the poly(A)(+) RNAs, see e.g. [1, 22, 23] and the references therein. Computational studies [12, 16, 17] showed that many of these "mRNA-like" ncRNAs, including prominent examples such as H19 and Xist, contain local, conserved secondary features, which hint at a functional role for the RNA itself.
Intriguingly, there is growing evidence of polyadenylation also for classical RNAs. The Gene Ontology has consequently been recently extended by the term GO:0043629 "ncRNA polyadenylation". Several examples come from yeast [24, 25], polyadenylated snRNAs have been observed in Dictyostelium discoideum [26, 27].
Recent studies identified a second nuclear poly(A) polymerase in yeast that is conserved through eukaryotes and tags aberrant non-coding RNAs for degradation . In addition, a growing number of microarray surveys of different organisms, in which oligo(dT) oligonucleotides have been used to amplify the cDNA probes, have also yielded hybridization signals from ncRNA targets. Examples of this are whole-genome (tiling) microarray experiments using RNA from human , mouse  and arabidopsis . Even though this method does not provide any direct evidence for polyadenylation, it indicates that the phenomenon of polyadenylated ncRNAs is more widespread than previously anticipated . Therefore, it is of interest to search for both ncRNAs that are normally not polyadenylated and mRNA-like ncRNAs in poly(A)(+) EST libraries.
Here, we report of the computational detection of novel ncRNAs and structural RNA cis-regulating (UTR) elements in the EST libaries of the Sino-Danish pig genome project . An automated pipeline EST2ncRNA was designed and applied to the assembled porcine EST data, which consists of 48,629 clusters (contigs) and 73,171 single reads (singletons). Contigs and single reads are collectively referred to as conreads  in the following. Predicted RNA structure candidates are further merged with expression information available in the PigEST resource  which contains an assembly of more than one million EST sequences. One-third of these originates from public available cDNA libraries and two-thirds originate from one normalized and 97 non-normalized cDNA libraries (35 tissues), of which 24 stem from developmental stages.
Known and novel ncRNAs and UTR elements in PigEST data
Known by sequence
Known by structure
Known by sequence
Known by structure
Sum of all RNA structures
The PigEST sequences contain 18,621 protein-coding RNAs which were detected directly by sequence similarity to protein databases, see . Protein-coding mRNAs are expressed mostly in large quantities, hence 80% of them form contigs. With the exception of SRA1  there are no known structured RNAs within ORFs, hence we removed the ORF regions. Nevertheless, the coding conreads were searched for cis-acting structured RNA elements in up- and downstream UTRs. The remaining ORF-free conreads are potential candidates for ncRNAs as well as UTR elements.
Homology search in mammals and few more distant species
The similarity search revealed that 27,578 contigs and 18,033 singeltons are at least partially but uniquely conserved on 71,112 loci in the cow (see methods), so far the most closely related mammal for which a genome has been sequenced. The conserved conreads consist of one-third of porcine ORF-free conreads (15,374 contigs, 15,552 singletons) and 80 percent of the protein-coding transcripts (12,204 contigs and 2,481 singletons). The remaining 76,189 conreads have no homologous sequence in the cow genome, hence they could not be further analysed by comparative genomics. This large amount of unaligned assembled ESTs could represent low quality singletons, transcriptional background or pig specific transcripts. At least in part, this large number is most likely an artifact since the current assembly of the bovine genome does not cover the entire genome . For 15,773 pig conreads, we observe split alignments mapping to different loci on the cow genome [see Additional file 1, Figure S3 for the number of loci per conread]. 77% of these map to the same chromosome in the same reading direction as one would expect for regularly spliced transcripts [see Additional file 1, Table S3]. The remaining cases are either EST sequencing artifacts, assembly problems in the current release of the cow genome or in the PigESTs. Conceivably some of them are exceptional transcripts such as the ones described in .
We then searched for similarities of the pig conreads with the 16 other vertebrates aligned to cow in the UCSC Genome Browser . We started from the 71,112 cow loci that we previously identified as homologous to a pig conread and considered both the pairwise blastz alignments  and the multiz alignments [see Additional file 1, Table S4] provided through the Genome Browser . We found for 66,350 loci a similar sequence in at least one additional species. The investigation of cow-human (65,196 homologous loci) and cow-mouse (55,416 homologous loci) pairwise alignments revealed that significantly more PigEST orthologs exist to human. This agrees with the already published thesis that the pig sequence space is closer to human than mouse . In the 4,227 pig-cow specific mappings there are only 177 protein-coding conreads (264 loci) included, which supports the lower evolutionary pressure of ORFs as well as their higher conservation. The mappings can be accessed online via the PigEST website .
Known non-coding RNAs and cis-acting RNA elements
The RaveNnA  scan using covariance models of known RNA structures revealed additional 54 contigs and 32 singletons matching 33 structures in Rfam v7.0  [see Additional file 1, Table S2). Of these, 44 contigs and 16 singletons are cis-acting RNA elements, of which 34 known RNA structures are located in protein-coding transcripts and 26 in ORF-free conreads. Again, the most frequently detected ncRNAs are microRNAs, snoRNAs and snRNAs. The total list of known cis-acting regulatory elements contains 37 Selenocysteine insertion sequences (SECIS), 15 Histone 3'-UTR stem-loops, and 8 Iron response elements. Additional tRNAs are detected in 34 contigs and 67 singletons through structure similarity. Approximately 100 tRNA candidates were successfully verified by tRNAscan_SE . Half of these were predicted as pseudogenes which are unusual tRNA homologues. Some functional ncRNAs derive from tRNAs. The most prominent example is the BC1 transcript in rodents , other tRNA/SINE-derived functional ncRNAs were recently described in . Our collection of expressed tRNA pseudogenes thus could contain novel tRNA-derived functional ncRNAs.
In summary, we obtain 137 known ncRNAs and cis-acting RNA elements by sequence similarity and structure similarity excluding tRNA candidates (see Table 1).
Candidates of novel non-coding RNAs and cis-acting (UTR) RNA elements
A control screen of the RNAz predictions using shuffled alignments as described in  yields an estimated False Discovery Rate (see methods for FDR calculation) of about 10% for ORF-free conreads. The CONC program, which uses an SVM to discriminate ORFs from other transcripts , did not provide additional information: CONC identified all RNAz predictions in ORF-free conreads as ncRNAs, which strengthens the rejection of RNAs with open reading frames (ORF) through the pipeline.
Reading direction of structured RNAs
The machine learning tool RNAstrand  classified 861 RNA structures to exist on the reverse complementary strand of non-protein coding conreads. In contrast, only 388 ones have a larger evidence to exist on the positive conread reading direction [see Additional file 1, Table S7]. A similar proportion of reading directions was observed for predicted RNA structures in coding conreads.
Mapping of structured RNAs to annotated UTRs
Predicted RNA structures, which are similar to known UTR regions, are counted as cis-acting (UTR) elements. Therefore, the more likely reading directions of RNAz predicted loci were scanned against the known gene annotation of human (hg17, May 2004) .
Of the RNA structures predicted in ORF-free conreads, 86% are conserved in human. We observed that 85% of the RNAz predictions are located far away from any known human gene (and UTR). These 1,004 conreads are labelled as putative ncRNAs in the porcine transcriptome. The remaining 15% (177 conreads) are homologous to human UTRs. Here, they are considered as putative cis-acting (UTR) RNA elements [see Additional file 1, Table S8]. Of these, 18 conreads contain a second RNA structure which is not similar to a human UTR and therefore they are also counted as ncRNA. The putative cis-regulatory (UTR) elements are located mostly on the sense strand, whereby the ratio of predicted ncRNAs on sense conreads is decreased to 29%. However, more than 35% of human conserved secondary RNA structures in pig are aligned to the reverse complementary strand of an annotated human UTR.
In addition to these we also investigated all the coding conreads, and we predicted high confidence RNA structures in 99 of these (95 contigs, 4 singletons), which do not overlap codon sequences, but are similar to a human UTR [see Additional file 1, Table S8]. These are around 40% (82 loci) of sense RNA structures as well as only 3% (18 loci) of antisense RNA structures of the human conserved coding conreads, which comprise 84% of all RNA structures in coding conreads. They are labelled as putative UTR elements together with the 177 ORF-free conreads described above. In total, 82% of the cis-acting (UTR) RNA structures are predicted in the positive conread reading direction. All high confidence putative ncRNAs and cis-acting RNA elements are available as the PigEST-ncRNA webserver .
The other human conserved RNA structures predicted on the positive strand of coding conreads are putative unspliced intronic ncRNAs or false positives. The predictions in the negative reading direction could be further candidates of independent transcriptional units. With slightly more relaxed RNAz criteria, the corresponding numbers are roughly 1,800 ORF-free conreads and 1,500 protein-coding transcripts which comprise predicted RNA structures in regions similar to known human UTRs, as well as around 9,700 ORF-free conreads with putative ncRNAs.
To summarise, high confidence novel RNA structures are predicted in 1,262 conreads. Together with 137 known RNA structures, the PigEST resource comprises trans-acting elements (ncRNAs) in 715 contigs and 340 singletons as well as cis-acting elements (inside UTRs) in 311 contigs and 51 singletons, of which 18 conreads contain both predictions of trans- and cis-acting elements (see Table 1).
Comparison with other screens for structured RNAs
Of the RNA structures which are conserved in human, 22% (199 loci) of the ncRNA candidates and 27% (77 loci) of UTR elements overlap (coverage >50%) a prediction from  or . These previously predicted structures occur in 270 conreads. On the other hand, the porcine transcriptome comprises 809 conreads with predicted ncRNAs located far away from a human ORF and 201 conreads with a cis-acting element which have not been reported before.
Transcriptional evidence in other organisms
To check for transcriptional evidence in other organisms, we searched in the NCBI dbEST  database and obtained a huge amount of hits. Similarities have been noticed in 73% of the RNAz predictions to ESTs discovered in other organisms than pig. High significant hits with an identity >95% and RNA structure coverage >80% have been detected for 30% of the conreads.
RNA structure predictions in known functional RNAs
The sequences which were matched initially by sequence and/or structure similarity to known functional RNAs listed in ncRNA databases, also are passed through the rest of the pipeline as a control. RNA structures are predicted by RNAz in six microRNAs and four cis-regulatory elements. The latter are confirmed through their homology to human UTR regions [see Additional file 1, Table S10]. In eight cases the genomic location of the known RNA and the prediction coincide. With more relaxed blastn options, described in the methods part, we found five additional annotated RNA structures (3 snoRNAs, 2 cis-acting elements) which were predicted by RNAz. Nevertheless, there are 16 conreads being detected as functional RNAs due to sequence similarity, which are not conserved in cow (Figure 2) and could not be verified by comparative genomic approaches.
Mapping the ncRNA candidates onto the PigEST cDNA Libraries
Using the expression information data from the PigEST resource  we inferred expression of the 715 contigs and 340 singletons containing ncRNA predictions or matches against Rfam (by sequence or structure similarity), where tRNA predictions were discarded due to apparent pseudogene annotation in the set. We conducted expression analysis of the corresponding conreads containing ncRNA predictions. The PigEST resource contains in total 92 useful non-normalized cDNA libraries from which expression patterns can be extracted. The expression of a contig in a cDNA library is simply counted as the fraction of EST reads from that library which are assembled into the contig.
We have implemented a pipeline to detect known and novel evolutionary conserved ncRNAs in assembled EST data through a combination of several stand-alone bioinformatic tools. As well as making ncRNA predictions, it also detects protein-coding RNAs and cis-acting regulatory regions (in UTRs) of mRNAs. We detected a large number of evolutionary conserved thermodynamically stable RNA structures in both ORF-free and protein-coding conreads. These conreads are plausible candidates for novel mRNA-like (polyadenylated) ncRNAs, many of which are spliced. The candidate set does not contain, on the other hand, large amount of intronic, poorly conserved, or non-structured ncRNA.
Protein-coding mRNAs are expressed on average at a much higher level compared to non-coding transcripts. As a consequence we observe that mRNAs, in contrast to ORF-free ESTs, are predominantly located in contigs rather than singletons. A high level of conservation of protein-coding sequences between pig conreads and the bovine genome emphasizes the similarities between the mammalian mRNA complements. In contrast, a higher rate of ncRNAs was predicted in singletons. These ncRNAs are probably expressed at low level which is also observed through the mapping to the individual PigEST cDNA libraries. The sequence conservation with other species is also much less pronounced for the ORF-free conreads. These observations are consistent with the idea that the non-coding parts of the transcriptome are much more variable between tissues and species .
In our data, the predicted RNA structures can be associated with protein coding genes either because the RNAz signal is located on a protein-coding conread or because an ORF-free conread shows significant sequence homology with a human UTR. The fraction of RNA structures in both groups is almost the same, hence they are biologically indistinguishable. One possible reason for the high abundance of UTR conreads that do not also contain ORFs is that UTRs with extensive secondary structures are larger and/or have a higher probability to give rise to truncated ESTs.
The pipeline could further be improved by including methods such as CMfinder  which can search for RNA structures in multiple sequences with low sequence similarity. Such an approach has shown to supplement methods like RNAz when the sequence similarity is so low that it affects the quality of multiple alignments made from methods based solely on sequence similarity (Torarinsson et al: Comparative genomics beyond sequence based alignments: RNA structures in the ENCODE regions, submitted). The expression analysis of the ncRNA candidates shows that most of these transcripts are cell-type specific. This observation might in part be due to the insufficifent cDNA library sizes. However, we obtained a four fold higher number of predictions in contigs than in singletons, supporting the cell specifity. In further contrast, more than 100 ncRNA candidates are expressed in at least ten cDNA libraries indicating transcription beyond the noise level. Strikingly the putative developmental regulatory ncRNAs are expressed at the highest level, which in agreement with earlier genomic analyses in other mammals [22, 61, 62]. Despite the strong computational confidence, only laboratory work can give a final verification.
In conclusion, the computational screen of the PigEST resource reported here provides strong confidence for a large number of conserved secondary structure elements in polyadenylated transcribed RNAs. The low expression levels of the predicted RNA candidates together with the observations that a larger relative fraction of them is found in cDNA libraries from developmental stages is in good agreement with the current paradigm of ncRNA roles in higher organisms and supports the thesis of functional polyadenylated ncRNAs. Since these seem to appear in low number in developmental tissues, this in itself provide a plausible explanation of why they previously have been overlooked.
The EST2ncRNA pipeline, presented in Figure 1, predicts ncRNA candidates by removal of all ORF regions from the source conreads. The pipeline compares all conreads to a closely related species for which a sequence is available, in this case the cow. This genome is then used as reference for homology searches in further organisms and for the construction of multiple alignments. These are screened in the next step for evolutionary conserved and thermodynamically stable structured ncRNAs using RNAz . At this stage, another algorithm could readily substitute RNAz. Predicted RNA structures that can be aligned with an UTR of another organism are labelled as "cis-regulatory" (UTR) elements. RNA structures in protein-coding conreads, which are not similar to an annotated UTR, are not further considered.
Data from the PigEST resource
The PigEST resource is based on 1,021,891 porcine EST sequences from which 636,516 were extracted from the Sino-Danish PigEST resource  and 385,375 from GenBank . The Sino-Danish PigEST resource originates from one normalized and 97 non-normalized cDNA libraries representing 35 different tissues and three developmental stages. The sequences were assembled by distiller  resulting in 48.629 contigs and 73,171 singletons. Protein-coding RNAs were searched by sequence similarity to the protein databases NCBI nonredundant and UniProt . As pragmatic selection criteria were applied an identity > 60% and known protein sequence coverage > 50% .
Sequence and structure similarity to known ncRNA families
The following ncRNA databases were scanned for primary sequence and secondary structure similarities: RNAdb (August 2004), Rfam (release 7.0, March 2005), miRBase (release 8.1, May 2006), fantom3_noncoding (release 3.0). The local blastn  searches were performed with standard parameters and selection thresholds of E-value < 10-20, identity >95% and subject coverage >85%.
The Rfam 7.0 covariance models, which represent known ncRNA families, were scanned by RaveNnA  to find structural similarity. The parameters (local or global alignment, window length) for each Rfam model were taken from the annotation file of the Rfam seed alignments . Since RaveNnA produce many low score hits, the candidates were further filtered. Hits were therefore ignored if (i) the matched conread regions included gaps, (ii) the modelled subsequences were shorter than 60% of the model window length or (iii) the number of modelled basepairs was not in a range of 20% around the basepair number of the model. Transfer RNAs were searched by tRNAscan_SE version 1.23 with default parameters, which combines tRNA covariance models with several heuristics.
Functional RNA elements in protein-coding conreads not overlapping the coding sequence were considered as cis-acting (UTR) RNA elements in the subsequent analysis.
The source pool was mapped against the genome of a closely related organism. The closest pig related even-toed ungulates (Artiodactyla) with an existing genomic sequence project  is the bovine Bos taurus (Btau-2.0, June 2005). Two parameter settings of blastn were applied for aligning EST data to genomic data, which were the standard parameter and EST specific parameter allocation recommended for annotation of genomic DNA with ESTs . The latter variant was realised through an adapted serial blast strategy  consisting of three steps to overcome the time problem: (i) a first blastn search with standard parameters, (ii) aligned conreads were retrieved, (iii) a second blastn search with EST specific parameters for the retrieved conreads against their related chromosomes. The result part is based on the blastn hits with EST specific parameter allocation. The non-overlapping blast hits with lowest E-value were filtered as conserved conreads if their E-value was less than 10-20, their length greater than 100 nt and a 100% identical sequence of at least a length of 75 nt exist. Conreads, which were not conserved in the closely related organism, were rejected from the pipeline and were stored as putative source organism specific transcript candidates.
More similar sequences to the conserved conreads in other organisms were detected by available genome-wide pairwise or multiple alignments from the UCSC genome browser  of an organism which was already aligned to the assembled ESTs. The pairwise alignments of cow (bosTau2, Mar. 2005, Baylor Btau 2.0) to human (hg17, May 2004) and mouse (mm7, Aug. 2005, NCBI Build 35) were available as chain files and were scanned by the UCSC liftOver tool  with a conread coverage ratio of 0.8. This implies that at least 80% of the reference organism subsequence had to match the alignment. The multiple alignment of human (hg17) to 11 mammals, three actinopterygii, one amphibia and one aves, generated with multiz, was available as maf file and was scanned with a conread coverage ratio of 0.6. The applied ratio was smaller for multiple alignments due to typically shorter alignments. To obtain multiple alignments containing the porcine conreads, we realigned the pig sequences to corresponding UCSC alignments by clustalw .
Prediction of conserved stable secondary structures
The alignments had to be pre-processed before RNAz 1.0(pre-release)  could be used to search for thermodynamically stable and conserved secondary RNA structures. In addition to several cleaning steps, which are described in the RNAz manual , the rnazWindows.pl tool sliced the alignments in 40 nt overlapping windows of size 120. This allows RNAz to find local structures. The pre-processed alignments were scored with RNAz using standard parameters plus the -both-strands parameter to score also the reverse complementary alignments. All alignments with classification score p > 0.5 were stored as conserved secondary RNA structures. The overlapping windows in the positive RNAz predictions were combined into clusters (loci) without attention for their strand predictions by the rnazCluster.pl script, which is also part of the RNAz package. RNAz predictions with RNAz classificator p > 0.9 and high thermodynamical stability described by a z-score < -3 were interpreted as high confidence hits.
The confidence of the predictions was measured by running RNAz again with randomized alignments. Therefore, the positions in the alignments of the preprocessed maf files were mononucleotidely shuffled with the RNAz tool rnazRandomizedAln.pl. The program aims to remove any correlation arising from a natural secondary structure while preserving mean pairwise identity and base composition. The false discovery rate was calculated as number of random hits related to native hits.
EST specific blastn parameters
The standard parameters of blastn are optimized for short alignments with a high identity and a short execution time. However, we were interested in a high conread coverage during the search for conread homologous genome sequences. Thereby, a small decrease of identity is justified due to the low quality of EST sequences and the structure conserved mutations in ncRNAs. The standard parameter allocation was compared with the one recommended for noncoding queries  and the one recommended for genomic DNA annotation with ESTs. The best balance between expected alignment length and percent identity, which are calculated by the High-scoring Segment Pair (HSP) of the blast algorithm, was provided by the EST specific parameters [see Additional file 1, Table S1].
The alignments of the PigEST data to the cow genomic data were provided by EST specific blastn parameters and standard parameters. Alignments generated with EST specific parameters were generally longer [see Additional file 1, Figure S1], which is more appropriate to find homologs to entire RNA transcripts. In general, EST specific alignments resulted in more positive RNAz predictions and the most structural RNA elements were predicted from both alignment sets [see Additional file 1, Figure S2]. Nevertheless, one third of positive predictions based on standard alignments were not detected by the other variant. These ESTs include possibly highly conserved short structures.
NcRNA candidates as cis-regulatory elements
The more reliable conread strand of each predicted conserved RNA structure (locus) was identified by RNAstrand v1.1.0 , which, like RNAz, RNAmicro and CONC, is based on a support vector machine (SVM) classifier. The prepared alignments of the related windows of each locus were merged and applied as input to the already trained SVM. The human gene annotation was used to identify RNA structures conserved in UTRs. RNAz hits are putative cis-acting elements, if their homologous human sequences overlap the UTRs of known genes, which are annotated in the knownGenes table (hg17, May 2004) of the UCSC genome browser . As mapping criteria was used an overlap of at least one base of the human homologs to a human UTR, whereas the RNAstrand predicted locus strand was applied. By cis-acting element candidates we considered only those matching known UTR regions, even though they can also occur outside of UTRs.
Putative novel miRNA precursors were detected by RNAmicro v1.1  in both reading directions of the alignments of positive RNAz predictions with window sizes 70, 100 and 130 nt and step size of 5 nt. RNAmicro assumes that input alignments include consensus secondary structures, wherefore the alignments of positive RNAz predictions were used. P-values > 0.9 were stored as putative microRNAs, whereas overlapping windows on the same strand were combined. Like RNAz results, RNAmicro predictions were verified by shuffled input alignments.
CONC v1.0 , a tool that predicts whether a sequence is protein-coding or not, was applied on all RNAz predicted loci of ORF-free conreads to verify them. To this end all 6 reading frames were investigated. Additional sequence similarities of the predicted RNA structures to known ncRNAs in the ncRNA databases were searched by blastn with more relaxed criteria, being an identity > 85% and a subject coverage > 60%. Furthermore, the high confidence ncRNA predictions were compared with the NCBI dbEST  database by standard blastn parameters and E-value < 10-10.
Availability and requirements
Perl source code of the 'EST2ncRNA' framework and the corresponding documentation are available for download from http://pigest.ku.dk/more/ncrna. At the same website we also provide the predicted ncRNAs and cis-regulatory elements. The comparion of Porcine and Bovine ESTs can be found at http://pigest.ku.dk/more/genomemap.
We thank Jana Hertel, Karsten Scheibye-Knudsen and Merete Fredholm for stimulating discussions. We thank Elfar Torarinsson for assistance with database and webserser setup. This work was supported in part by the German DFG Bioinformatics Initiative, DFG SPP "Deep Metazoan Phylogeny" (STA 850/2-1), the Danish Research Council for Technology and Production (FTP), the Lundbeck foundation and the Danish Center for Scientific Computation.
- 2.Cawley S, Bekiranov S, Ng HH, Kapranov P, Sekinger EA, Kampa D, Piccolboni A, Sementchenko V, Cheng J, Williams AJ, Wheeler R, Wong B, Drenkow J, Yamanaka M, Patel S, Brubaker S, Tammana H, Helt G, Struhl K, Gingeras TR: Unbiased mapping of transcription factor binding sites along human chromosomes 21 and 22 points to widespread regulation of noncoding RNAs. Cell. 2004, 116: 499-509. 10.1016/S0092-8674(04)00127-8.PubMedCrossRefGoogle Scholar
- 3.Carninci P, Kasukawa T, Katayama S, Gough J, Frith MC, Maeda N, Oyama R, Ravasi T, Lenhard B, Wells C, FANTOM Consortium; RIKEN Genome Exploration Research Group and Genome Science Group (Genome Network Project Core Group): The transcriptional landscape of the mammalian genome. Science. 2005, 309: 1559-1563. 10.1126/science.1112014.PubMedCrossRefGoogle Scholar
- 5.Manak JR, Dike S, Sementchenko V, Kapranov P, Biemar F, Long J, Cheng J, Bell I, Ghosh S, Piccolboni A, Gingeras TR: Biological function of unannotated transcription during the early development of Drosophila melanogaster. Nat Genet. 2006, 38 (10): 1151-1158. 10.1038/ng1875.PubMedCrossRefGoogle Scholar
- 13.Havgaard JH, Torarinsson E, Gorodkin J: Fast Pairwise Structural RNA Alignments by Pruning of the Dynamical Programming Matrix. In revision. 2007Google Scholar
- 14.Missal K, Rose D, Stadler PF: Non-coding RNAs in Ciona intestinalis. Bioinformatics. 2005, 21 (S2): i77-i78.Google Scholar
- 17.Pedersen JS, Bejerano G, Siepel A, Rosenbloom K, Lindblad-Toh K, Lander ES, Kent J, Miller W, Haussler D: Identification and classification of conserved RNA secondary structures in the human genome. PLoS Comput Biol. 2006, 2 (4): e33-10.1371/journal.pcbi.0020033.PubMedCentralPubMedCrossRefGoogle Scholar
- 18.Torarinsson E, Sawera M, Havgaard J, Fredholm M, Gorodkin J: Thousands of corresponding human an mouse genomic regions unalignable in primary sequece contain common RNA structure. Genome Res. 2006, 16: 885-889. 10.1101/gr.5226606. [Erratum: Genome Res. 16: 1439 (2006)]PubMedCentralPubMedCrossRefGoogle Scholar
- 19.Washietl S, Pedersen JS, Korbel JO, Gruber A, Hackermüller J, Hertel J, Lindemeyer M, Reiche K, Stocsits C, Tanzer A, Ucla C, Wyss C, Antonarakis SE, Denoeud F, Lagarde J, Drenkow J, Kapranov P, Gingeras TR, Guigó R, Snyder M, Gerstein MB, Reymond A, Hofacker IL, Stadler PF: Structured RNAs in the ENCODE Selected Regions of the Human Genome. Genome Res. 2007, 17: 852-864. 10.1101/gr.5650707.PubMedCentralPubMedCrossRefGoogle Scholar
- 20.Rose D, Hackermüller J, Washietl S, Findeiß S, Reiche K, Hertel J, Stadler PF, Prohaska SJ: Computational RNomics of Drosophilids. BMC Genomics. 2007, [Accepted]Google Scholar
- 22.Ravasi T, Suzuki H, Pang KC, Katayama S, Furuno M, Okunishi R, Fukuda S, Ru K, Frith MC, Gongora MM, Grimmond SM, Hume DA, Hayashizaki Y, Mattick JS: Experimental validation of the regulated expression of large numbers of non-coding RNAs from the mouse genome. Genome Res. 2006, 16 (1): 11-19. 10.1101/gr.4200206.PubMedCentralPubMedCrossRefGoogle Scholar
- 26.Hinas A, Larsson P, Avesson L, Kirsebom LA, Virtanen A, Soderbom F: Identification of the major spliceosomal RNAs in Dictyostelium discoideum reveals developmentally regulated U2 variants and polyadenylated snRNAs. Eukaryot Cell. 2006, 5 (6): 924-934. 10.1128/EC.00065-06.PubMedCentralPubMedCrossRefGoogle Scholar
- 30.Stolc V, Samanta MP, Tongprasit W, Sethi H, Liang S, Nelson DC, Hegeman A, Nelson C, Rancour D, Bednarek S, Ulrich EL, Zhao Q, Wrobel RL, Newman CS, Fox BG, Phillips GNJ, Markley JL, Sussman MR: Identification of transcribed sequences in Arabidopsis thaliana by using high-resolution genome tiling arrays. Proc Natl Acad Sci USA. 2005, 102 (12): 4453-4458. 10.1073/pnas.0408203102.PubMedCentralPubMedCrossRefGoogle Scholar
- 32.Gorodkin J, Cirera S, Hedegaard J, Gilchrist MJ, Panitz F, Jørgensen CB, Scheibye-Knudsen K, Arvin T, Lumholdt S, Sawera M, Green T, Nielsen BJ, Havgaard JH, Wang J, Li H, Li R, Liu B, Hu S, Dong W, Li W, Yu J, Wang J, Stærfeldt HH, Madsen LB, Thomsen B, Hornshøj H, Bujie Z, Wang X, Wang X, Bolund L, Brunak S, Yang H, C B, Fredholm M: Porcine transcriptome analysis based on 97 non-normalized cDNA libraries and assembly of 1,021,891 expressed sequence tags. Genome Biol. 2007, 8 (4): R45-10.1186/gb-2007-8-4-r45.PubMedCentralPubMedCrossRefGoogle Scholar
- 33.PigEST resource 1.0. [http://pigest.ku.dk]
- 34.Hube F, Guo J, Chooniedass-Kothari S, Cooper C, Hamedani MK, Dibrov AA, Blanchard AA, Wang X, Deng G, Myal Y, Leygue E: Alternative splicing of the first intron of the steroid receptor RNA activator (SRA) participates in the generation of coding and noncoding RNA isoforms in breast cancer cell lines. DNA Cell Biol. 2006, 25 (7): 418-428. 10.1089/dna.2006.25.418.PubMedCrossRefGoogle Scholar
- 35.Bovine Genome Project. [http://www.hgsc.bcm.tmc.edu/projects/bovine/]
- 38.Genome Browser website. [http://genome.ucsc.edu/]
- 39.Wernersson R, Schierup MH, Jørgensen FG, Gorodkin J, Panitz F, Stærfeldt HH, Christensen OF, Mailund T, Hornshoj H, Klein A, Wang J, Liu B, Hu S, Dong W, Li W, Wong GK, Yu J, Wang J, Bendixen C, Fredholm M, Brunak S, Yang H, Bolund L: Pigs in sequence space: A 0.66x coverage pig genome survey based on shotgun sequencing. BMC Genomics. 2005, 6: 70-10.1186/1471-2164-6-70.PubMedCentralPubMedCrossRefGoogle Scholar
- 40.PigEST genomemap. [http://pigest.ku.dk/more/genomemap]
- 41.Griffiths-Jones S, Moxon S, Marshall M, Khanna A, Eddy SR, Bateman A: Rfam: annotating non-coding RNAs in complete genomes. Nucleic Acids Res. 2005, D121-124. 33 DatabaseGoogle Scholar
- 42.Pang KC, Stephen S, Engstrom PG, Tajul-Arifin K, Chen W, Wahlestedt C, Lenhard B, Hayashizaki Y, Mattick JS: RNAdb-a comprehensive mammalian noncoding RNA database. Nucleic Acids Res. 2005, D125-130. 33 DatabaseGoogle Scholar
- 43.FANTOM3. [http://fantom.gsc.riken.go.jp/]
- 44.Griffiths-Jones S, Grocock RJ, van Dongen S, Bateman A, Enright AJ: miRBase: microRNA sequences, targets and gene nomenclature. Nucleic Acids Res. 2006, D140-144. 10.1093/nar/gkj112. 34 DatabaseGoogle Scholar
- 52.PigEST ncRNA. [http://pigest.ku.dk/more/ncrna]
- 57.Yelin R, Dahary D, Sorek R, Levanon EY, Goldstein O, Shoshan A, Diber A, Biton S, Tamir Y, Khosravi R, Nemzer S, Pinner E, Walach S, Bernstein J, Savitsky K, Rotman G: Widespread occurrence of antisense transcription in the human genome. Nat Biotechnol. 2003, 21 (4): 379-386. 10.1038/nbt808.PubMedCrossRefGoogle Scholar
- 61.Cheng J, Kapranov P, Drenkow J, Dike S, Brubaker S, Patel S, Long J, Stern D, Tammana H, Helt G, Sementchenko V, Piccolboni A, Bekiranov S, Bailey DK, Ganesh M, Ghosh S, Bell I, Gerhard DS, Gingeras TR: Transcriptional maps of 10 human chromosomes at 5-nucleotide resolution. Science. 2005, 308 (5725): 1149-1154. 10.1126/science.1108625.PubMedCrossRefGoogle Scholar
- 63.Benson DA, Karsch-Mizrachi I, Lipman DJ, Ostell J, Wheeler DL: Genbank. Nucleic Acids Res. 2006, D16-20. 10.1093/nar/gkj157. 34 DatabaseGoogle Scholar
- 67.Rfam Home Page. [http://www.sanger.ac.uk/Software/Rfam/]
- 68.I, Yandell M, Bedell J: BLAST O'Reilly. 2003Google Scholar
- 69.I: Serial BLAST searching. Bioinformatics. 2003, 19 (12): 1492-1496. 10.1093/bioinformatics/btg199.Google Scholar
- 70.Lift genome annotations. [http://genome.ucsc.edu/cgi-bin/hgLiftOver]
- 71.Thompson JD, Higgins DG, Gibson TJ: CLUSTAL W: improving the sensitivity of progressive multiple sequence alignment through sequence weighting, position specific gap penalties and weight matrix choice. Nucleic Acids Res. 1994, 22 (22): 4673-4680. 10.1093/nar/22.22.4673.PubMedCentralPubMedCrossRefGoogle Scholar
- 72.RNAz manual. [http://www.tbi.univie.ac.at/~wash/RNAz/]
- 73.Blastn parameters for noncoding queries. [http://stevemount.outfoxing.com/Posting0004.html]
- 74.Genome Browser Custom Track. [http://genome.ucsc.edu]
- 76.R Project. [http://www.r-project.org]
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.