- Methodology article
- Open Access
Genetic validation of whole-transcriptome sequencing for mapping expression affected by cis-regulatory variation
- Tomas Babak1,
- Philip Garrett-Engele1,
- Christopher D Armour2, 6,
- Christopher K Raymond2, 6,
- Mark P Keller3,
- Ronghua Chen1,
- Carol A Rohl1,
- Jason M Johnson1,
- Alan D Attie3,
- Hunter B Fraser†2, 4Email author and
- Eric E Schadt†2, 5Email author
© Babak et al; licensee BioMed Central Ltd. 2010
- Received: 26 May 2010
- Accepted: 13 August 2010
- Published: 13 August 2010
Identifying associations between genotypes and gene expression levels using microarrays has enabled systematic interrogation of regulatory variation underlying complex phenotypes. This approach has vast potential for functional characterization of disease states, but its prohibitive cost, given hundreds to thousands of individual samples from populations have to be genotyped and expression profiled, has limited its widespread application.
Here we demonstrate that genomic regions with allele-specific expression (ASE) detected by sequencing cDNA are highly enriched for cis- acting expression quantitative trait loci (cis- eQTL) identified by profiling of 500 animals in parallel, with up to 90% agreement on the allele that is preferentially expressed. We also observed widespread noncoding and antisense ASE and identified several allele-specific alternative splicing variants.
Monitoring ASE by sequencing cDNA from as little as one sample is a practical alternative to expression genetics for mapping cis-acting variation that regulates RNA transcription and processing.
- Quantitative Trait Locus
- Antisense Transcription
- BTBR Mouse
- Allelic Bias
- Additive Quantitative Trait Locus Effect
The genetics of genome-wide gene expression has emerged as an important new field with potential to transform our understanding of a broad scope of topics, ranging from basic regulation of transcription to mechanisms of complex human diseases. In most studies of gene expression genetics, genetically diverse individuals are genotyped at genetic markers that characterize most of the common DNA variation in the population and are also phenotyped by measuring the abundances of thousands of mRNA transcripts . These molecular phenotypes are then genetically mapped like any other quantitative trait, revealing quantitative trait loci (QTL) where a polymorphism affects a transcript's abundance levels . These studies have led to the construction of regulatory networks that are predictive of disease states such as obesity in both mouse and human [2, 3]. They have also shown that gene expression QTL (eQTL) in humans can uncover the mechanisms of action of disease-associated SNPs previously implicated by genome-wide association studies (GWAS) [4, 5], and have even implicated new SNPs as additional disease-associated loci . At the heart of all eQTL studies is polymorphic gene expression. This can be caused by genetic variants that act in cis or in trans. A major difference between the two classes is that because a cis-acting allele acts only on the chromosomal copy on which it resides, a heterozygous cis-acting polymorphism results in allele-specific effects, such as higher expression of allele A vs allele B, even though both alleles are present in the same nucleus and thus experience the same trans- acting environment . In contrast, a heterozygous trans- acting polymorphism cannot lead to allele-specific effects, because it does not differentiate between the two alleles.
Most studies of gene expression genetics have classified expression variation within species as due to either cis- or trans- acting factors; this is typically done by genetically mapping variation in expression levels, and inferring that a gene whose expression maps very close to the gene itself is most likely influenced by one or more cis-acting variants [1, 6]. Cis-acting polymorphisms have been found to be extremely common in all studies published to date, and generally exert much stronger effects on gene expression than do trans- acting polymorphisms. For example, recent studies of human gene expression genetics have inferred between 85-100% of significant eQTLs to be cis-acting [2, 5], although our power to detect these two types of events has not been exhaustively compared. trans- acting eQTL may eventually be found to be more prevalent, once larger sample sizes and thus greater power are achieved, and they have also been found to be useful in the context of coexpression networks and causal inference [2, 3, 5].
It has been appreciated for many years--since well before the first studies of genome-wide gene expression genetics--that measuring ASE can reveal the presence of heterozygous cis-acting polymorphisms. In fact, ASE is a necessary consequence of heterozygous cis-acting polymorphisms. ASE can be measured by quantifying the abundance of each allele of a transcribed heterozygous polymorphism (such as a SNP) among a gene's transcripts. Since the genomic ratio of heterozygous alleles in a diploid is 1:1, any significant deviation from this among mRNA transcripts suggests allele-specificity. Primarily due to technical limitations, measuring ASE has not been widely applied to identify cis-acting polymorphisms. The emergence of high-throughput cDNA sequencing (i.e. RNA/NSR-seq) has enabled whole-genome identification of ASE and has been applied to mapping imprinted loci [7, 8]. In addition, two targeted approaches employing PCR  and padlock-capture [10, 11] have shown that up to 25% of genes may be preferentially expressed from one allele and that a significant proportion of these are tissue-specific .
Although it is assumed that identifying ASE by high-throughput sequencing is an alternative to mapping cis-eQTL by microarrays, this has not been shown experimentally. cis-eQTL result in higher expression of one allele over the other in any given sample. This bias consistently favors one parental allele if the causal polymorphism is in linkage disequilibrium (LD) with the transcribed SNPs used to quantify ASE. In this study, we validate cDNA sequencing toward identifying regions affected by inheritable cis-acting regulatory variants by demonstrating extensive overlap between ASE and cis-eQTL.
Using a custom Agilent murine microarray , we profiled 500 adipose and islet samples from an F2 intercross population constructed from the BTBR and C57BL/6J strains on an Ob null background (referred to here as the BTBRxB6 cross), and detected cis-eQTL for expression traits using a standard regression procedure [12, 13]. We excluded 1,291 genes from this analysis in which probes overlapped a known  or predicted (see Methods) BTBR/B6 SNP from our analysis, since these can lead to false-positive cis-eQTL by impacting hybridization kinetics. At LOD > 3 (FDR = 0.01) we detected 3,367 and 3,819 cis-eQTL genes (of 34,257 on array) in adipose and islets respectively which we used as baseline lists for comparison.
Under the null-hypothesis (no allelic bias), the ratio of BTBR to B6 reads reflects the allelic proportions within the RNA pool. Although the average allelic ratio was 50:50, we used the microarray genotyping information to augment the precision of our analysis by exactly defining the ratio at each locus (see Methods). 2,230 adipose and 1,444 islets genes met our criteria for reliably measuring ASE, most notably containing a minimum of 10 allele-specific sequencing reads covering at least three SNPs (see Methods). 719 and 501 genes respectively were also detected in cis-eQTL which by itself represents a significant overlap (Fisher Exact Test p < 1e-11).
ASE identified by NSR-seq agrees with cis-eQTL identified by microarrays
Previous sequencing-based ASE studies used F1 s [7, 8] where both copies of each allele are always present at exactly equal ratios, thus enabling a simple statistical analysis of ASE. Our primary goal was validation of the method. Using the same samples for identifying both ASE and cis-eQTL was thus essential for avoiding inter-sample biases and additional effects induced by the exceptionally high heterozygosity of F1 s. Nonetheless, we observed strong overlaps in F1 s as well (Additional File 1, Figure S4), demonstrating that using single samples is sufficient to map regions affected by cis-acting variation and that artificial F1 effects are not extensive.
A number of explanations may account for any disagreement on the direction of bias between the NSR-seq and microarray data. Most obvious are the technical differences in determining transcriptional abundance: microarrays monitor 3'UTR abundance whereas NSR-seq captures the entire transcript. Any transcriptional processing effect not reflected in the 3'UTR may thus lead to an overall difference, and any 3'UTR processing effect will be exaggerated on arrays. Furthermore, SNPs in microarray probe regions may lead to artifactual differences in BTBR and B6 transcript level measurements, and false-positive SNP predictions that artifactually bias sequencing allele counts toward B6 (the reference genome). To gain insight into these discrepancies, we Sanger-resequenced SNPs in all genes that disagreed in their direction of bias between the two technologies at high ASE and cis-eQTL thresholds in adipose (n = 25 adipose samples, |log(binomial-p)| > 3, |add. eff.| > 0.05). Strain bias in 7 genes could not be sufficiently distinguished by analyzing trace files. Of the remaining 18 genes, 15 (83%) agreed in direction with NSR-seq, of which 5 (28%) had no detectable BTBR signal, presumably due to a false-positive SNP, and 3 (17%) agreed with microarray-inferred strain bias (Additional File 1, Table S1). This suggests that measuring ASE by high-throughput sequencing may be less susceptible to artifacts than mapping cis-eQTLs in an F2 population.
ASE is widespread and encompasses noncoding RNA
Comparing the complete observed and expected distributions of allele-specificity (see Methods), we found that 36.7% of the BTBR SNPs covered by at least 10 reads in adipose showed greater allele-specificity than expected, and at 100× coverage, this increased to 59.9%. In the islet data, allelic bias was even more common: 42.0% of SNPs were biased at 10× coverage, and 68.0% were biased at 100×. At 100× coverage, we can detect 1.70-fold differences at p = 0.01 with 50% power (or at p = 0.05 with 82% power); at 1000× coverage, this drops to 1.18-fold at p = 0.01 with 50% power (or 78% power at p = 0.05). Unfortunately we cannot estimate the total extent of ASE, since this strongly depends on the distribution of effect sizes for cis-eQTL (e.g., having many weak effect sizes below our current detection threshold would indicate there is still much more ASE to be found). Future studies with higher coverage of the transcriptome will be able to address this issue.
Antisense transcription occurs more frequently from the same allele vs. independent alleles
Widespread antisense transcription in mammals is well documented [19–21], although a general function for these transcripts has not been established. Recently, several groups have identified clustering of short antisense transcripts immediately upstream of transcription start sites (TSSs) [22, 23]. In yeast, these short transcripts are immediate targets of the exosome suggesting that they are non-functional derivatives of an intrinsically bidirectional RNA polymerase II . The strand-specific nature of our RNA amplification protocol allowed us to further explore this by further dissecting antisense transcription by their allelic origin. In order to maximize the number of assayable sites with antisense transcription, we used previously published data from F1 B6xCAST/Ei embryos , which have on average >4-fold higher SNP density over B6/BTBR . We have also done these analyses with BTBRxB6 data and achieved similar, but as expected, weaker trends resulting from lower sensitivity (BTBRxB6 analyses shown in Additional File 1, Figure S6).
Identification of allele-specific splicing events
Allele-specific splicing candidates.
Exon Included Reads (B6:CAST)
Exon Skipped Reads (B6:CAST)
p(ASE) is less than*
Gene function (Gene Ontology)
Conserved protein domain encoded by skipped exon (Pfam)
reduction of cytosolic calcium ion concentration
No domain reported
Bromodomain adj. to zinc finger domain (PTHR22880)
No domain reported
cortical actin cytoskeleton organization
4.1 C-terminal domain (PF05902)
polynucleotide adenylyltransferase activity
PAP/OAS1 substrate-binding domain (SSF81631))
RNA splicing; body morphogenesis
DNA2/NAM7 helicase family member (PTHR10887:SF5)
Ankyrin repeat (SSF48403)
nuclear migration; nervous system development
CS (PS51203), HSP20-like chaperones (SSF49764)
mRNA processing, RNA binding
Splicing factor 45 (PTHR13288:SF9)
We found a highly significant overlap in cis-eQTL genes identified by microarray profiling and ASE genes identified by NSR-seq. Improving overlap with increasing additive QTL effect sizes and/or LBP demonstrates that, as expected, both approaches reliably detect allele-specific expression despite a number of key differences in the technologies. For example, the majority of microarray gene-expression platforms, including the platform used here, rely on 3'-biased amplification protocols and thus position microarray probes near the 3'-end of the gene. With NSR-seq we monitored the entire transcript, including introns, which we previously found to improve sensitivity  presumably because most intronic reads correspond to unprocessed pre-mRNA or degradation products. Any allele-specific events outside the microarray probe region will thus skew the comparison. Unknown SNPs can also cause disagreement: 1) they can lead to artifactual genetic associations if they are within probe regions , 2) can bias allelic representation in NSR-seq if located within priming sites, and 3) can lead to a bias towards aligning NSR-seq reads to the reference genome (though this last scenario is expected to be rare, since it requires having both a known and an unknown SNP in the same read; consistent with this expectation, visual inspection of Figure 3 reveals little bias). False-positive SNPs accounted for 28% of extreme cases where NSR-seq disagreed with microarrays. Although this is a substantial overestimate of total effect since selection of these genes was biased toward strong disagreement, it highlights the importance of high-quality SNP maps for both methods.
While our results mostly agree with those from more traditional approaches using hundreds of microarrays to measure gene expression among F2 mice, there are several distinct advantages of the NSR-seq approach. First, NSR-seq results are not limited by a fixed set of probes, and because of this we were able to find allele-specific instances of splicing and antisense transcription that were invisible to our microarrays. Second, identification of cis- eQTL does not depend on arbitrary genomic distance cutoffs with NSR-seq, in contrast to microarray studies which will inevitably misclassify some trans-eQTL as cis, and vice versa. Third, the NSR-seq approach can be applied to any outbred diploid or polyploid species, even those for which microarrays are not readily available. Finally, NSR-seq can be applied to a single F1 individual, with a single sequencing run costing several thousand dollars, as opposed to applying microarrays to each of hundreds of F2 individuals, saving a great deal of time and expense (though pooling F2 individuals may more closely agree with cis-eQTL from microarrays, if genetic interactions in the F2 population are not captured in the F1). On the other hand, the greatest disadvantage of NSR-seq is its inability to detect trans-eQTL. Measuring ASE in the context applied here where animals were pooled at the RNA level cannot be used to map the genomic region that contains the causal effect. Sequencing or microarray profiling many samples will always be required for mapping QTL. A second disadvantage is that in outbred species with short LD blocks (such as human), the causal cis-eQTL polymorphism will often not be in LD with any transcribed SNPs, and in these cases pooling will not reveal ASE. Third, pooling RNA from different individuals could introduce biases. These can be minimized by keeping track of the mass of RNA from each individual or by pooling by equal mass. It is also possible that one or a few individuals in the pool have significantly different expression levels and contribute unbiasly to the ASE signal. We show that this is very unlikely given our strong agreement on ASE in F1 biological replicates (Figure 1a), but ruling it out for the entire pool would require testing individuals. Nevertheless, pooling human samples has already been shown to reveal many cis-eQTL . To detect cis-eQTL where the causal polymorphism is not in LD with any transcribed SNPs, NSR-seq can be performed on individual samples, as long as genotype phasing is known . A method able to quickly and efficiently identify cis-eQTL genome-wide will find applications in many areas. The dependence of cis-eQTL on environmental conditions--a subject not previously studied, in large part due to its prohibitive cost--can now be studied efficiently and comprehensively. The method can be applied to hybrids between distinct species (as has already been done with low-throughput pyrosequencing for Drosophila hybrids) to reveal all cis-acting gene expression differences, and inform us of the importance of cis-regulation in evolution. Since the action of positive selection can now be inferred solely from cis-eQTL (H. Fraser; personal communication), selection on gene expression can be measured by NSR/RNA-seq in a wide range of species. And finally, cis-acting polymorphisms have been shown to be highly enriched for SNPs associated with human disease risk in genome-wide association studies, so compiling catalogs of genes affected by cis-eQTL in various tissues, populations, and disease states could be extremely useful for inferring which disease associations are likely due to cis-acting effects on gene expression, and even more importantly, which genes are perturbed by the disease-associated variants (see Additional File 1, Supplement).
Validated against expression genetics, cDNA sequencing is an effective strategy for identifying allele-specific expression which can be used to map inheritable cis-acting variation. Application across multiple samples has potential to yield insight into polygenic phenotypes including complex disease.
Sample collection and microarray analysis
The BTBR × B6 F2 mice were constructed by intercrossing F1 animals obtained by crossing C57BL/6 (B6) BTBR mice carrying the leptinob/ob (ob) mutation. The resulting F2 animals were housed in an environmentally-controlled facility on a 12 hr light/dark cycle (6 AM - 6 PM, respectively). Mice were provided free access to water at all times and to a standard rodent chow (Purina #5008) ad libitum, except during a fasting period (8 AM - noon) in order to obtain plasma at 10 weeks of age, after which they were sacrificed by decapitation. For each animal the right gonadal fat pad (adipose) and pancreas were collected for expression profiling. The adipose tissues were flash frozen in liquid nitrogen. Intact pancreatic islets were isolated from the F2 mice using a collagenase digestion procedure as previously described . A detailed description of islet isolation, RNA purification, and microarray analyses is available in Additional File 1 (Supplementary Methods). All animal handling procedures were approved by University of Wisconsin Animal Care and Use Committee.
Total RNA from 100 adipose, islet, liver, and hypothalamus samples (see above) was pooled and subjected to strand-specific, whole-cell NSR-seq . Libraries were sent to Illumina (Hayward, California) for single-end 36 nt sequencing for total depth of 2 G/sample. Novoalign (Novocraft) was used to align against NCBI mouse genome release 36 (UCSC Feb. 2006 release [mm8]) and a collection of splice junctions generated from Refseq genes , ENSEMBL genes , and UCSC Known Genes . Predicted splice junctions from ESTs , Genscan , and N-scan predictions  were also considered in regions that lack coding gene models. All possible splice sites spanning up to two exon skipping events in gene/transcript models above were represented. A minimum of 5 nt overlap per flanking junction sequence was required for alignment to be considered, selected on basis of maximizing overall alignment sensitivity (data not shown). All reads that aligned uniquely to the genome or splice-sites, and redundantly mapped reads that overlap unique reads in only one genomic location, were retained for further analysis. 73,556,741/89,349,136 adipose, 89,418,898/108,429,611 islets, 81,359,243/98,250,529 hypothalamus, and 49,297,475/61,390,433 liver were successfully aligned by employing the above criteria. BTBR/B6 raw sequence data is available at NCBI Short Read Archive under accession SRA008619.3; previously published CAST/B6 data is accessible under SRA008621.10.
Quantification of allele-specific events from sequencing data
ASE was assessed by summing allele-specific reads that align over independently identified SNPs ; using the SNP to distinguish the allelic origin. For comparison with microarray transcript levels, allelic counts were summed across all SNPs within the transcript boundaries (including introns), provided that the sequencing reads were in the same orientation as the transcript. To exclude artifactual allelic bias arising from false-positive SNPs or differential priming events, we required a minimum of three SNPs within the gene and agreement on strain bias at the majority of all SNPs/transcript. We excluded SNPs where the average Illumina-phred score was below 20, since scores below 20 do not accurately reflect sequencing errors (Illumina, personal communication). We also excluded C/A, A/C, and G/T (B6/BTBR) SNPs since the sequence data had disproportional amounts of reference mismatches of these variety, indicative of high error rates.
where All_genes was the total number of genes on the microarray after removal of genes with SNPs in probe regions (n = 34,257).
Antisense analysis was conducted for SNPs where reads mapped to both strands. Since our NSR-seq approach may incorrectly detect up to 0.7% antisense reads , we used the binomial to exclude potential artifacts where expression is predominant from one strand (p < 1e-4; Bonferroni-corrected). LBP scores were normalized to the mean for each strand (Figure 4b) to correct for overlap arising from slightly higher numbers of B6 reads than BTBR (presumably from genomic alignment bias). Allele-specific splicing was assessed when a SNP overlaps reads that support an alternative splice junction (i.e. both isoforms are represented). Only reads with the same orientation as the spliced transcript were considered.
Experimental validation of allele-specific antisense events
0.5 ug total RNA pooled from four B6 × CAST and four CAST × B6 9.5 day-old embryos was either 1) reverse transcribed using Qiagen OneStep RT-PCR kit (Qiagen) according to the manufacturer's instructions with 35 rounds of PCR (for alternative splice-site amplification), or 2) reverse-transcribed with superscript III (Invitrogen), RNAse H treated (Invitrogen), and amplified by Roche high-Fidelity PCR (Roche; 35 cycles) according to the manufacturer's instructions (for strand-specific amplification of antisense transcription). Primer sequences are listed in Additional File 1, Table S2.
We would like to thank Brian DeVeale and Derek van der Kooy at the University of Toronto for donating all mouse embryos used in this study.
- Gibson G, Weir B: The quantitative genetics of transcription. Trends Genet. 2005, 21 (11): 616-623. 10.1016/j.tig.2005.08.010.PubMedView ArticleGoogle Scholar
- Chen Y, Zhu J, Lum PY, Yang X, Pinto S, MacNeil DJ, Zhang C, Lamb J, Edwards S, Sieberts SK: Variations in DNA elucidate molecular networks that cause disease. Nature. 2008, 452 (7186): 429-435. 10.1038/nature06757.PubMed CentralPubMedView ArticleGoogle Scholar
- Emilsson V, Thorleifsson G, Zhang B, Leonardson AS, Zink F, Zhu J, Carlson S, Helgason A, Walters GB, Gunnarsdottir S: Genetics of gene expression and its effect on disease. Nature. 2008, 452 (7186): 423-428. 10.1038/nature06758.PubMedView ArticleGoogle Scholar
- Fraser HB, Xie X: Common polymorphic transcript variation in human disease. Genome Res. 2009, 19 (4): 567-75. 10.1101/gr.083477.108.PubMedView ArticleGoogle Scholar
- Schadt EE, Molony C, Chudin E, Hao K, Yang X, Lum PY, Kasarskis A, Zhang B, Wang S, Suver C: Mapping the genetic architecture of gene expression in human liver. PLoS Biol. 2008, 6 (5): e107-10.1371/journal.pbio.0060107.PubMed CentralPubMedView ArticleGoogle Scholar
- Doss S, Schadt EE, Drake TA, Lusis AJ: Cis-acting expression quantitative trait loci in mice. Genome Res. 2005, 15 (5): 681-691. 10.1101/gr.3216905.PubMed CentralPubMedView ArticleGoogle Scholar
- Babak T, Deveale B, Armour C, Raymond C, Cleary MA, van der Kooy D, Johnson JM, Lim LP: Global survey of genomic imprinting by transcriptome sequencing. Curr Biol. 2008, 18 (22): 1735-1741. 10.1016/j.cub.2008.09.044.PubMedView ArticleGoogle Scholar
- Wang X, Sun Q, McGrath SD, Mardis ER, Soloway PD, Clark AG: Transcriptome-wide identification of novel imprinted genes in neonatal mouse brain. PloS one. 2008, 3 (12): e3839-10.1371/journal.pone.0003839.PubMed CentralPubMedView ArticleGoogle Scholar
- Main BJ, Bickel RD, McIntyre LM, Graze RM, Calabrese PP, Nuzhdin SV: Allele-specific expression assays using Solexa. BMC genomics. 2009, 10: 422-10.1186/1471-2164-10-422.PubMed CentralPubMedView ArticleGoogle Scholar
- Lee JH, Park IH, Gao Y, Li JB, Li Z, Daley GQ, Zhang K, Church GM: A robust approach to identifying tissue-specific gene expression regulatory variants using personalized human induced pluripotent stem cells. PLoS genetics. 2009, 5 (11): e1000718-10.1371/journal.pgen.1000718.PubMed CentralPubMedView ArticleGoogle Scholar
- Zhang K, Li JB, Gao Y, Egli D, Xie B, Deng J, Li Z, Lee JH, Aach J, Leproust EM: Digital RNA allelotyping reveals tissue-specific and allele-specific gene expression in human. Nat Methods. 2009, 6 (8): 613-618. 10.1038/nmeth.1357.PubMed CentralPubMedView ArticleGoogle Scholar
- Wang S, Yehya N, Schadt EE, Wang H, Drake TA, Lusis AJ: Genetic and genomic analysis of a fat mass trait with complex inheritance reveals marked sex specificity. PLoS genetics. 2006, 2 (2): e15-10.1371/journal.pgen.0020015.PubMed CentralPubMedView ArticleGoogle Scholar
- Haley CS, Knott SA: A simple regression method for mapping quantitative trait loci in line crosses using flanking markers. Heredity. 1992, 69 (4): 315-324.PubMedView ArticleGoogle Scholar
- Frazer KA, Eskin E, Kang HM, Bogue MA, Hinds DA, Beilharz EJ, Gupta RV, Montgomery J, Morenzoni MM, Nilsen GB: A sequence-based variation map of 8.27 million SNPs in inbred mouse strains. Nature. 2007, 448 (7157): 1050-1053. 10.1038/nature06067.PubMedView ArticleGoogle Scholar
- Armour CD, Castle JC, Chen R, Babak T, Loerch P, Jackson S, Shah JK, Dey J, Rohl CA, Johnson JM: Digital transcriptome profiling using selective hexamer priming for cDNA synthesis. Nat Methods. 2009, 6 (9): 647-9. 10.1038/nmeth.1360.PubMedView ArticleGoogle Scholar
- Ge B, Pokholok DK, Kwan T, Grundberg E, Morcos L, Verlaan DJ, Le J, Koka V, Lam KC, Gagne V: Global patterns of cis variation in human cells revealed by high-density allelic expression analysis. Nature genetics. 2009, 41 (11): 1216-1222. 10.1038/ng.473.PubMedView ArticleGoogle Scholar
- Verlaan DJ, Ge B, Grundberg E, Hoberman R, Lam KC, Koka V, Dias J, Gurd S, Martin NW, Mallmin H: Targeted screening of cis-regulatory variation in human haplotypes. Genome Res. 2009, 19 (1): 118-127. 10.1101/gr.084798.108.PubMed CentralPubMedView ArticleGoogle Scholar
- Mituyama T, Yamada K, Hattori E, Okida H, Ono Y, Terai G, Yoshizawa A, Komori T, Asai K: The Functional RNA Database 3.0: databases to support mining and annotation of functional RNAs. Nucleic acids research. 2009, D89-92. 10.1093/nar/gkn805. 37 DatabaseGoogle Scholar
- Katayama S, Tomaru Y, Kasukawa T, Waki K, Nakanishi M, Nakamura M, Nishida H, Yap CC, Suzuki M, Kawai J: Antisense transcription in the mammalian transcriptome. Science. 2005, 309 (5740): 1564-1566. 10.1126/science.1112009.PubMedView ArticleGoogle Scholar
- Yelin R, Dahary D, Sorek R, Levanon EY, Goldstein O, Shoshan A, Diber A, Biton S, Tamir Y, Khosravi R: Widespread occurrence of antisense transcription in the human genome. Nature biotechnology. 2003, 21 (4): 379-386. 10.1038/nbt808.PubMedView ArticleGoogle Scholar
- Okazaki Y, Furuno M, Kasukawa T, Adachi J, Bono H, Kondo S, Nikaido I, Osato N, Saito R, Suzuki H: Analysis of the mouse transcriptome based on functional annotation of 60,770 full-length cDNAs. Nature. 2002, 420 (6915): 563-573. 10.1038/nature01266.PubMedView ArticleGoogle Scholar
- He Y, Vogelstein B, Velculescu VE, Papadopoulos N, Kinzler KW: The antisense transcriptomes of human cells. Science. 2008, 322 (5909): 1855-1857. 10.1126/science.1163853.PubMed CentralPubMedView ArticleGoogle Scholar
- Seila AC, Calabrese JM, Levine SS, Yeo GW, Rahl PB, Flynn RA, Young RA, Sharp PA: Divergent transcription from active promoters. Science. 2008, 322 (5909): 1849-1851. 10.1126/science.1162253.PubMed CentralPubMedView ArticleGoogle Scholar
- Neil H, Malabat C, d'Aubenton-Carafa Y, Xu Z, Steinmetz LM, Jacquier A: Widespread bidirectional promoters are the major source of cryptic transcripts in yeast. Nature. 2009, 457 (7232): 1038-1042. 10.1038/nature07747.PubMedView ArticleGoogle Scholar
- Matlin AJ, Clark F, Smith CW: Understanding alternative splicing: towards a cellular code. Nature reviews. 2005, 6 (5): 386-398. 10.1038/nrm1645.PubMedView ArticleGoogle Scholar
- Kwan T, Benovoy D, Dias C, Gurd S, Provencher C, Beaulieu P, Hudson TJ, Sladek R, Majewski J: Genome-wide analysis of transcript isoform variation in humans. Nature genetics. 2008, 40 (2): 225-231. 10.1038/ng.2007.57.PubMedView ArticleGoogle Scholar
- Rabaglia ME, Gray-Keller MP, Frey BL, Shortreed MR, Smith LM, Attie AD: Alpha-Ketoisocaproate-induced hypersecretion of insulin by islets from diabetes-susceptible mice. American journal of physiology. 2005, 289 (2): E218-224.PubMedGoogle Scholar
- Pruitt KD, Tatusova T, Maglott DR: NCBI reference sequences (RefSeq): a curated non-redundant sequence database of genomes, transcripts and proteins. Nucleic Acids Res. 2007, D61-65. 10.1093/nar/gkl842. 35 DatabaseGoogle Scholar
- Hubbard TJ, Aken BL, Ayling S, Ballester B, Beal K, Bragin E, Brent S, Chen Y, Clapham P, Clarke L: Ensembl 2009. Nucleic Acids Res. 2009, D690-697. 10.1093/nar/gkn828. 37 DatabaseGoogle Scholar
- Kuhn RM, Karolchik D, Zweig AS, Wang T, Smith KE, Rosenbloom KR, Rhead B, Raney BJ, Pohl A, Pheasant M: The UCSC Genome Browser Database: update 2009. Nucleic Acids Res. 2009, D755-761. 10.1093/nar/gkn875. 37 DatabaseGoogle Scholar
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.