|
|
Proc Natl Acad Sci U S A. 2007 March 20; 104(12): 5020–5025. Published online 2007 March 12. doi: 10.1073/pnas.0611078104. | PMCID: PMC1820821 |
Copyright © 2007 by The National Academy of Sciences of the USA Genetics Evidence of spatially bound gene regulation in Mus musculus: Decreased gene expression proximal to microRNA genomic location Hidenori Inaoka,*† Yutaka Fukuoka,* and Isaac S. Kohane‡§¶ *School of Biomedical Science and †Institute of Biomaterials and Biomedical Engineering, Tokyo Medical and Dental University, Chiyoda-ku, Tokyo 101-0062, Japan; ‡Informatics Program, Children's Hospital, Center for Biomedical Informatics, and Partners Center for Genetics and Genomics, Harvard Medical School, Boston, MA 02115; and §Division of Health Sciences and Technology, Harvard University and Massachusetts Institute of Technology, Cambridge, MA 02139 Received August 29, 2006. Freely available online through the PNAS open access option. The extent, spatially and in time, of the phenomenon of localized decreased expression in the chromosomal vicinity of microRNA (miRNA) previously described in Caenorhabditis elegans is reproduced in Mus musculus across a wide range of tissues in several independent experiments. Computationally predicted miRNA targets are enriched in the vicinity of miRNAs, and transcription factors are identified as the class of genes that systematically exhibit this localized decrease. Also, those mRNA with AT-rich UTRs, particularly those that are not in the vicinity of CpG islands, most often exhibit this localized decrease. This localization broadens with the shift from developing to mature/differentiated tissues and suggests a developmentally controlled and spatially bound regulation. Keywords: cis-regulation, development, transcription The regulatory properties of colocation of genetic elements in chromosomal linear space ( 1, 2) and in the three-dimensional space of chromosomes and the nucleus ( 3) have become increasingly apparent in eukaryotes. Coexpression bias of proximal genes has been documented across a range organisms, from Saccharomyces cerevisiae (yeast) ( 4), Arabidopsis thaliana (mustard plant), Caenorhabditis elegans (worm) ( 5), Drosophila melanogaster (fly), Mus musculus (mouse), to Homo sapiens ( 6– 9). The role, if any, of noncoding RNAs and microRNAs (miRNAs) in particular in this localized control of expression has yet to be determined. miRNAs are central gene expression regulators, functioning as posttranscriptional suppressors ( 10, 11). In metazoa, these short noncoding RNAs induce translational repression, and mRNA decay by binding to target mRNAs and therefore affect mRNA levels directly ( 12– 15). Understanding of what determines the specificity of mRNA targeting by miRNA is incomplete: It is known that recognition of a target mRNA occurs in the context of a protein complex (RNA-induced silencing complex) that requires neither perfect sequence complementarity nor thermodynamic stability between the miRNA and its target. Two broad classes of miRNA-binding sites in animal mRNA have been identified: 5′ dominant and 3′ compensatory ( 16). Targets that are 5′ dominant have consecutive Watson–Crick base pairings with the 5′ end of the miRNA (often called the miRNA “seed”) and have lesser complementarity to the rest of the molecule. Targets that are 3′ compensatory have a weak 5′ base pairing and rely on strong compensatory pairing to the 3′ end of the miRNA. There is also mounting evidence that small noncoding RNAs have an important role in the control of the dynamics of localized gene expression through heterochromatin formation ( 17– 19). Furthermore, analysis of C. elegans expression data sets provides evidence that gene expression tends to be decreased locally in the neighborhood of miRNAs throughout the genome, at least in that organism ( 20). Consequently we have hypothesized that miRNAs may have a role in the localized expression of genes in their neighborhood in “higher” organisms, including mammals. As a preliminary exploration of this hypothesis, we have determined the extent, spatially along the chromosome and in developmental time, of the phenomenon of localized decreased expression in M. musculus across a wide range of tissues in independent experiments. We defined a series of increasingly large chromosomal “windows” centered on each documented miRNA location. For each tissue, the expression of mRNA within each successive window was normalized by the overall expression of mRNA in that tissue as measured by expression microarrays. We compared these levels with those calculated in windows selected from 49,800 random locations in the murine genome. Furthermore, we determined how the composition of the proximal genome might affect the observed expression levels, including the presence of CpG islands, AT enrichment of mRNA targets, and proximal distribution of mRNA targets of each miRNA. The phenomenon of lower expression in the neighborhood of miRNAs is marked (approximately two-fold decrease) and widespread. We found extensive evidence of localized decreased gene expression in the neighborhood of miRNAs in many murine developmental models, including eight tissue-specific developmental models (see Methods). Localized decreased expression was observed in normal mouse cerebellar development ( 21) and in the ptch mutant mouse ( 21), lung development ( 22), thymic T cell ( 23), preimplantation embryo development ( 24), oocyte development ( 25), and two models of skeletal muscle cell growth ( 26, 27). These decreases in expression were not found in the neighborhood of randomly selected chromosomal locations. This finding is illustrated in a, in which the average expression level across all of the developmental systems is plotted as a function of chromosomal distance to the miRNA in base pairs. As plotted by the solid line, the average expression of mRNA starts at ≈50% of the transcriptome-wide average at 10 4 bp from each miRNA and reaches the average by 10 5 bp. One SEM error bar is shown with triangles. On the same plot, the dashed line shows the average mRNA expression as a function of distance from 49,800 randomly chosen positions with one SEM error bar. This line is consistently flat across all distances from the random position in contrast to the same calculation with respect to miRNAs. Of the 349 genes that showed lower expression in at least two experiments from either developmental or mature tissue data sets (to limit the number of genes included whose expression was not systematically affected in the miRNA neighborhood) and are in a 400-kb window of any miRNA, the only molecular function category that showed enrichment by the Fischer exact test ( P < 0.0001) was transcription factor activity/DNA binding (see Methods). One hundred and fifty-four genes that had higher expression in two or more experiments were only enriched for translational initiation factor activity as shown in Table 1. The full list of genes is shown in supporting information (SI) Tables 2 and 3. To determine to what extent the genomic distribution of transcription factors might contribute to this finding, we determined which gene ontology (GO) ( 28) categories were overrepresented in the vicinity of miRNA, irrespective of gene expression level. Within 100 kb of miRNA, no category reached significance after Bonferroni correction, although the categories of nucleic acid binding/DNA binding and translation/translation initiation were top-ranked. The elimination of another possible confounder of these findings, miRNA clustering ( 29, 30), did not alter the above findings (see Methods). | Table 1.GO category of low- and high-expression genes in miRNA vicinity |
The gene expression profile around miRNA was repeated for mature tissues to determine whether the effect noted in development persisted. The analysis was repeated for the following mature tissues obtained in one set of experiments described by Su et al. ( 31): mouse myocardium, kidney, brain, liver, lung, skeletal muscle, spleen, and thymus. Again, localized decreased gene expression was noted in all eight tissues but was even broader than in the developing tissues. b plots the average mRNA expression with respect to miRNA position. As in a, expression levels return from their nadir near the miRNAs to the global average with increasing distance but here only at 10 6 bp, rather than 10 5 bp. As before, there is no such trend with expression data computed with respect to random chromosomal locations. To determine whether this difference between developmental and mature tissues persisted within data sets, we selected the earliest time points and the latest time points in the lung time series and repeated the same analysis showing the broadening trend (not reaching significance) as shown in . In this plot, the solid line illustrates the average expression of mRNA as a function of distance from miRNAs in the developing lung by using only the first two time points in the time series. The dashed line illustrates the same relationship but by using the last two points in the developmental time series. The early development (solid) line returns to the global average in a shorter distance than the late development (dashed) line. Because represents a small fraction of one of the many data sets used in the prior figures, the error bars are larger. A chromosomal view of the above miRNA-centric analyses is shown for a few chromosomes in (all chromosomes shown in SI Fig. 6). In these diagrams, the averaged expression across all tissues and across 100-kb windows centered on successive mRNA positions is plotted, and the positions of miRNA are indicated with arrows. It can be seen that the miRNAs are in or near local dips in this averaged expression. Also, there are many other dips in locations other than the currently known miRNA locations, although we know from the prior analysis that randomly chosen locations will not show the consistent colocation with expression nadirs seen for miRNA. To determine whether targeting of mRNA transcripts by the local/ cis miRNA might account for the observed decreased local expression, the number of putative mRNA targets for each miRNA was calculated in windows ranging from 10 kb to 200 mb centered on the miRNA. By computationally predicting targets (TargetscanS; ref. 32), the enrichment of targets in the larger windows was found to be lower than in the more proximal windows when normalized for gene count per window. The peak enrichment for targets as calculated by TargetscanS appears between 100 and 400 kb with a steady decrease to 15% of the peak value. The trend was not observed with the random location-centered data and is illustrated in , which shows the average across all of the miRNAs of the number of computationally predicted mRNA targets of the specific miRNA in whose vicinity the mRNA are found. The number of computed targets is expressed as a percentage of potential targets to normalize for the differing numbers of genes with increasing distance from each miRNA. The downwards trend in target enrichment from 400 kb (smaller windows have two or less predicted targets) to 200 mb is highly significant as calculated by the Cochran–Armitage test for trends in linear proportions ( 33, 34) (χ 2 of 131.2 and P < 2.2E-16). DNA methylation at CpG-rich sites is a component of one of the well documented mechanisms of transcriptional silencing ( 35). If the decreased expression levels in the neighborhood of miRNA were mediated in part by this mechanism, then we would expect a difference in the expression profiles in the vicinity of those miRNA in CpG-rich chromosomal regions versus those that were CpG-poor. The developmental data sets were reanalyzed by splitting them into two roughly equally sized sets: those miRNA That had at least one CpG island identified in silico within 100 kb (130 miRNA) and those without such an island (109 miRNA). As shown in , there is indeed a significant difference between the local expression profiles. Both profiles show locally decreased expression near miRNA locations, but without a CpG island in proximity to the miRNA (downward-pointing triangles), the phenomenon is much more pronounced and over a larger window in contrast to the expression level with proximal CpG islands (up to 10 6 bp as compared with between 10 4 and 10 5 bp). Recent analyses ( 36) suggest that genes containing AT-rich UTRs tend to be more involved in transcription and translation processes whereas genes with GC-rich UTRs are more often involved in signal transduction and translation modification, and the majority of miRNA targets are AT-rich. For those genes that were near miRNA and have a proximal CpG island, the 3′ UTRs showed less AT enrichment (51.51 ± 1.16% vs. 56.6 ± 1.62%) than those without a proximal CpG island (as previously defined), and are therefore less likely to be directly targeted by the neighboring miRNA. However, there was no difference in AT enrichment of the 3′ UTR between high- (52.59 ± 0.42%) and low- (52.02 ± 0.33%) expressed genes in the vicinity of miRNA. This result suggests that the CpG effect illustrated in is more than merely a proxy for AT richness. To determine whether these observations were confounded by the distribution of CpG islands around miRNA, we compared the distribution of CpG islands in a 100-kb window around microRNA vs. the distribution around randomly selected mRNA. There was no significant difference in the two distributions except that in miRNA the CpG peak is downstream, whereas in mRNA it is slightly upstream ( SI Fig. 7). In our broadening understanding of the role of small RNAs in genetic regulation, their role in gene silencing and particularly epigenetic regulation, including the initiation of heterochromatin formation has only recently become apparent ( 18, 19). We provide here evidence of a modest but systematic and widespread decreased expression of coding gene mRNA that is localized to and centered on the position of miRNA throughout the murine genome. Furthermore, this localized decrease is robustly reproduced. It is noted both in late development (i.e., after apparent tissue fate determination) in eight independent experiments and eight different mature differentiated tissues. The extent of this effect appears to extend to ≈10 5 bp in the developing tissues but tends to have a larger extent (approximately10 6 bp) in the differentiated tissues. This trend is reproducible within single-tissue developmental time series. These results are consistent with earlier results described in the worm transcriptome ( 20). This localized decreased expression effect is a subtle one in that several genes are expressed at high levels in proximity to miRNAs. It only becomes apparent across thousands of genes, which speaks to the dominance of other mechanisms. Nevertheless, the persistence of this effect genome-wide does suggest a robust control mechanism. The observation that the only functional category of gene that is statistically enriched in this localized effect is that of DNA-binding proteins or transcription factors suggests that the primary consequence of this phenomenon is indirect, through the modulation of the expression of the regulators. Across the aforementioned multiple independent experiments, this hypothesis is further supported by the highly significant enrichment of miRNA targets within the vicinity of each miRNA that are specific to that miRNA. Due to the relatively low specificity and sensitivity of current computational predictors of miRNA targets, the quantification of this enrichment can only be tentative at present. Given the role of miRNA in epigenetic silencing ( 17), the interaction between CpG islands, gene expression, development and proximity to miRNAs that we observe here is perhaps not surprising even if the genome-wide nature of this phenomenon is previously unreported. The observed increased AT richness of 3′ UTRs of the genes in the miRNA neighborhoods without proximal CpG islands vs. those that do have proximal CpG islands is consistent with the observation described in ref. 36 of transcriptional and translational function of the genes with such AT-rich UTRs. Nonetheless, as noted above, AT enrichment alone does not identify those genes that are likely to have lower expression levels in the vicinity of a miRNA. Unaccounted for is the enrichment for translation initiation factors among those genes up-regulated in the vicinity of miRNA. Whether this phenomenon is a primary effect of miRNA targeting or a secondary regulatory effect remains to be determined. Nonetheless, this observation is intriguing because many of these genes are members of the Argonaute family. Argonaute genes have been implicated in miRNA-mediated translation repression and miRNA-directed mRNA degradation and chromatin modification ( 14, 37, 38). As to the function of this large-scale organization of miRNA-centered domains of decreased expression increasing with maturation, two possibilities suggest themselves. First, beyond the initiation of heterochromatin formation, miRNA may be involved (as are other noncoding RNAs; refs. 39 and 40) in the extension of heterochromatic domains as the developmental program progresses ( 41). Second, as chromatin unwinds to allow miRNA expression, neighboring transcription factors (TFs) may have coevolved 3′ UTR targets for these miRNA so as to avoid large-scale trans-acting effects of these TFs across the genome in the absence of specific TF enhancers. Determination of how these localized effects change with experimentally modified levels of specific miRNA levels may be illuminating in this regard. The position pi of each miRNA documented in the murine genome was obtained from the Sanger database ( http://microrna.sanger.ac.uk/sequences/ftp.shtml). Gene expression within each data set was mean and unit SD normalized. Successively larger windows wj of length l were centered on each pi. All of the coding genes gk within wj were identified, and the average expression of all of the gk within wj was calculated and expressed as a quotient with respect to the expression of all measured genes in a particular data set (“normalized expression level”). For comparison, an equal number of windows of length l were identified centered on randomly picked locations in the mouse genome, and the same window-specific average expression levels were calculated. For this purpose, 49,800 random locations were picked. For the chromosomal plots of , sliding windows of size 100 kb were calculated, each centered on each successive gene on a chromosome. The average of the logarithm base 10 of the expression of the genes in that window was then plotted at the chromosomal location corresponding to the center of each window. The data sets include the following from the Gene Expression Omnibus ( 42): GDS237 and GDS257; thymic T cell development, GDS813; preimplantation embryo development, GSE3351; gene expression during oocyte development, GSE3787; C2C12 mouse skeletal muscle cells grown in differentiation medium for 5 days, GSE989 and GSE990; myogenic differentiation; and GDS182, a collection of tissues. Also the ptch mutant and cerebellar development data sets are available upon request, and the lung development data set is available at http://lungtranscriptome.bwh.harvard.edu/exp.data.html. Gene Category Enrichment. For each data set, the expression levels were normalized by the overall average. The genes in a 400-kb window were extracted as the subject of the analysis. Genes with a normalized value of ≥1.1 were categorized as increased. Similarly, genes with a normalized value of ≤0.9 were regarded as decreased. Then for each gene, the number of conditions (across all of the eight developmental tissue data sets), in which the gene was increased (or decreased), was counted. The enrichment of each category of gene annotation was calculated by using the GOHyperG program in Bioconductor ( 43) and with the microarray platform used for the expression studies as the source of background annotation frequencies. GOHyperG calculates the P values from the hypergeometric distribution for the GO categories on the basis of the background frequency of each GO category on a specified expression microarray. This procedure is equivalent to using Fisher's exact test ( 44). mRNA Targets in miRNA Vicinities. miRNA targets were computationally identified by using target-scanning software TargetScanS ( 32) to find all predicted mRNA targets within a specified wj of pi. Percentages of targets with respect to the total number of mRNA (potential targets) in each wj were graphed in . CpG Island Determination. We downloaded a list of CpG islands (file seq_1020;cpg_islands.md.gz) from the National Center for Biotechnology Information web site ( ftp.ncbi.nlm.nih.gov/genomes/M_musculus/mapview). We used CpG islands which met the “strict” conditions of ( i) 500-bp minimum length, ( ii) ≥50% GC content, and ( iii) ≥0.60 observed CpG/expected CpG ( 45). The list includes the chromosomal position of each CpG island (if known). We then calculated a distance between every possible combination of CpG island and miRNA on the same chromosome. Because approximately half of the miRNAs have at least one CpG island within 100-kb proximity, we considered an miRNA that has a CpG island located <100 kb away as an miRNA with CpG and the others as miRNAs without CpG. The average normalized expression level of the genes near each group of miRNAs was calculated in differently sized windows for the two groups. Results are plotted in . miRNA Clusters. Clusters of miRNA were defined as contiguous groups of miRNAs in which no miRNA is more distant than 10 kb from another miRNA. The average expression levels were recalculated across the same successive windows as before but with and without the miRNA clusters. As shown in SI Fig. 8, there was no significant difference in the locally decreased pattern expression of coding genes around clustered vs. unclustered miRNA. We thank Prof. Simon Kasif, Prof. Joseph Majzoub, Prof. Louis Kunkel, Alal Eran, and the reviewers for valuable suggestions for improving the manuscript. I.S.K. was supported in part by the National Institutes of Health National Center for Biomedical Computing Grant 5U54LM008748-02. | | GO | gene ontology | | | miRNA | microRNA. |
1. Caron H, van Schaik B, van der Mee M, Baas F, Riggins G, van Sluis P, Hermus MC, van Asperen R, Boon K, Voute PA, et al. Science. 2001;291:1289–1292. [PubMed] 2. Lawrence JG. Cell. 2002;110:407–413. [PubMed] 3. Stein GS, Lian JB, Montecino M, Stein JL, van Wijnen AJ, Javed A, Pratap J, Choi J, Zaidi SK, Gutierrez S, et al. Chromosome Res. 2003;11:527–536. [PubMed] 4. Cohen BA, Mitra RD, Hughes JD, Church GM. Nat Genet. 2000;26:183–186. [PubMed] 5. Roy PJ, Stuart JM, Lund J, Kim SK. Nature. 2002;418:975–979. [PubMed] 6. Lee JM, Sonnhammer EL. Genome Res. 2003;13:875–882. [PubMed] 7. Megy K, Audic S, Claverie JM. Genome Biol. 2003;4:P1. [PubMed] 8. Fukuoka Y, Inaoka H, Kohane IS. BMC Genomics. 2004;5:4. [PubMed] 9. Thygesen HH, Zwinderman AH. BMC Bioinformatics. 2005;6:10. [PubMed] 10. Bartel DP. Cell. 2004;116:281–297. [PubMed] 11. Ambros V. Nature. 2004;431:350–355. [PubMed] 12. Wu L, Fan J, Belasco JG. Proc Natl Acad Sci USA. 2006;103:4034–4039. [PubMed] 13. Bagga S, Bracht J, Hunter S, Massirer K, Holtz J, Eachus R, Pasquinelli AE. Cell. 2005;122:553–563. [PubMed] 14. Jing Q, Huang S, Guth S, Zarubin T, Motoyama A, Chen J, Di Padova F, Lin SC, Gram H, Han J. Cell. 2005;120:623–634. [PubMed] 15. Wu L, Belasco JG. Mol Cell Biol. 2005;25:9198–9208. [PubMed] 16. Brennecke J, Stark A, Russell RB, Cohen SM. PLoS Biol. 2005;3:e85. [PubMed] 17. Bernstein E, Allis CD. Genes Dev. 2005;19:1635–1655. [PubMed] 18. Verdel A, Jia S, Gerber S, Sugiyama T, Gygi S, Grewal SI, Moazed D. Science. 2004;303:672–676. [PubMed] 19. Grewal SI, Moazed D. Science. 2003;301:798–802. [PubMed] 20. Inaoka H, Fukuoka Y, Kohane IS. BMC Bioinformatics. 2006;7:112. [PubMed] 21. Kho AT, Zhao Q, Cai Z, Butte AJ, Kim JY, Pomeroy SL, Rowitch DH, Kohane IS. Genes Dev. 2004;18:629–640. [PubMed] 22. Mariani TJ, Reed JJ, Shapiro SD. Am J Respir Cell Mol Biol. 2002;26:541–548. [PubMed] 23. Hoffmann R, Bruno L, Seidl T, Rolink A, Melchers F. J Immunol. 2003;170:1339–1353. [PubMed] 24. Zeng F, Baldwin DA, Schultz RM. Dev Biol. 2004;272:483–496. [PubMed] 25. Pan H, O'Brien M, J., Wigglesworth K, Eppig JJ, Schultz RM. Dev Biol. 2005;286:493–506. [PubMed] 26. Tomczak KK, Marinescu VD, Ramoni MF, Sanoudou D, Montanaro F, Han M, Kunkel LM, Kohane IS, Beggs AH. FASEB J. 2004;18:403–405. [PubMed] 28. Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, Cherry JM, Davis AP, Dolinski K, Dwight SS, Eppig JT, et al. Nat Genet. 2000;25:25–29. [PubMed] 29. Seitz H, Royo H, Bortolin ML, Lin SP, Ferguson-Smith AC, Cavaille J. Genome Res. 2004;14:1741–1748. [PubMed] 30. Tanzer A, Stadler PF. J Mol Biol. 2004;339:327–335. [PubMed] 31. Su AI, Cooke MP, Ching KA, Hakak Y, Walker JR, Wiltshire T, Orth AP, Vega RG, Sapinoso LM, Moqrich A, et al. Proc Natl Acad Sci USA. 2002;99:4465–4470. [PubMed] 32. Lewis BP, Burge CB, Bartel DP. Cell. 2005;120:15–20. [PubMed] 33. Cochran WG. Biometrics. 1954;10:417–451. 34. Armitage P. Biometrics. 1955;11:375–386. 35. Fahrner JA, Baylin SB. Genes Dev. 2003;17:1805–1812. [PubMed] 36. Robins H, Press WH. Proc Natl Acad Sci USA. 2005;102:15557–15562. [PubMed] 37. Zilberman D, Cao X, Jacobsen SE. Science. 2003;299:716–719. [PubMed] 38. Meister G, Landthaler M, Peters L, Chen PY, Urlaub H, Luhrmann R, Tuschl T. Curr Biol. 2005;15:2149–2155. [PubMed] 39. Petrie VJ, Wuitschick JD, Givens CD, Kosinski AM, Partridge JF. Mol Cell Biol. 2005;25:2331–2346. [PubMed] 40. Chow JC, Yen Z, Ziesche SM, Brown CJ. Annu Rev Genomics Hum Genet. 2005;6:69–92. [PubMed] 41. Eshed Y, Bowman JL. Dev Cell. 2004;7:629–630. [PubMed] 42. Barrett T, Suzek TO, Troup DB, Wilhite SE, Ngau WC, Ledoux P, Rudnev D, Lash AE, Fujibuchi W, Edgar R. Nucleic Acids Res. 2005;33:D562–D566. [PubMed] 43. Gentleman RC, Carey VJ, Bates DM, Bolstad B, Dettling M, Dudoit S, Ellis B, Gautier L, Ge Y, Gentry J, et al. Genome Biol. 2004;5:R80. [PubMed] 44. Adams WT, Skopek TR. J Mol Biol. 1987;194:391–396. [PubMed] 45. Takai D, Jones PA. Proc Natl Acad Sci USA. 2002;99:3740–3745. [PubMed]
|
