《Topology of the human and mouse m6A RNA methylomes revealed by m6A-seq.pdf》由会员分享,可在线阅读,更多相关《Topology of the human and mouse m6A RNA methylomes revealed by m6A-seq.pdf(8页珍藏版)》请在三一文库上搜索。
1、ARTICLE doi:10.1038/nature11112 Topology of the human and mouse m6A RNA methylomes revealed by m6A-seq Dan Dominissini1,2*, Sharon Moshitch-Moshkovitz1*, Schraga Schwartz3*, Mali Salmon-Divon1, Lior Ungar2,4, Sivan Osenberg1,2, Karen Cesarkas1, Jasmine Jacob-Hirsch1, Ninette Amariglio1, Martin Kupie
2、c4, Rotem Sorek3 transcript architecture is shown beneath, with thin parts corresponding to UTRs and thicker ones to CDS; exonexon junctions are indicated by vertical black lines. a.u., arbitrary units. a b c AAAAAAAAA Sequencing primer Fragmentation (100 nt) IP with anti-m6A ab RNA sample Random pr
3、imed cDNA library generation, adaptor ligation and Illumina sequencing Input control AAAAAAAAA A T m m6A NHCH3 N N N N O OHOH OH P O OCH2HO A T m AAAAAAAAA A T m A T m A T m A T m A T m A T m m6A Control 15 10 Density (103) 5 0 0100200 A T m Bits 2 1 0 2 1 m6A 1 23 4 Peak to consensus distance (nt)
4、Figure 1|m6A-seq capture of modified RNA fragments exposes an enriched motif. a, Schematic diagram of the m6A-seq protocol. ab, antibody; nt, nucleotide. IP, immunoprecipitation. b, Sequence logo representing the deduced consensus motif followingclusteringofallenriched motifs. c, Density plot illust
5、rating the distribution of distance between the peaks of m6A/control fragments and the nearest m6A consensus motif (RRACU). RESEARCH ARTICLE 2 0 2 |N A T U R E| V O L 4 8 5 |1 0M A Y2 0 1 2 Macmillan Publishers Limited. All rights reserved2012 landmarks in their architecture, based on the single-tra
6、nscript, intron-free transcriptome data set (to avoid misinterpretations due to peaks falling into different isoform-dependent landmarks). We noted that m6A peaks were markedly correlated with two distinct coordinates: immediately following the transcription start site (TSS) and in the vicinity of t
7、he stop codon (Fig. 2a). Overall, stop codon peaks were more pronounced than TSS peaks. To assess the enrich- ment methodically, we assigned each m6A peak to one of five non- overlapping transcript segments: TSS; 59 untranslated region (UTR); coding sequence (CDS); stop codon; and 39 UTR (Fig. 2b, p
8、ie chart), andthennormalizedbytherelativefractioneachsegmentoccupiedin thetranscriptome(Fig.2b,histogram).Thestopcodonsegment(400- nucleotide window centred on the stop codon) stood out as most enriched in m6A peaks, with 28% of the peaks, representing a ,2.6- foldenrichmentoverthedistributionexpect
9、edbychance (P,2.23102308, x2test; Fig. 2e and Supplementary Fig. 5). Moreover, 33.5%ofall adequately expressed genes (thatis, .40reads per kilobase (kb) had a peak in this region. In only 2.3% of the cases didamethylationconsensusformatthestopcodon(incorporatingits two consecutive purines), indicati
10、ng that methylation typically did not occur precisely at the stop codon but rather in its vicinity. Nonetheless, the much longer CDS segment, although slightly depleted in m6A peaks when normalized by length, harboured the largest fraction of peaks (37%) (Fig. 2b, e). Notably, these peaks tend to oc
11、cur within unusually long internal exons: of 2,838 m6A peaks within internal exons, 2,453 (87%) are in exons longer than 400 nucleotides, compared to only 14% of the negative control peaks (P53.43102156, t-test) (Fig. 2c, Supplementary Fig. 6 and Supplementary Table 1). Comparing the lengths of flan
12、king introns andthestrengthsofthe59and39splicesitesandpolypyrimidinetract, we found only minor differences between long methylated exons and their unmethylated counterparts, which for the most part were not statistically significant. In contrast to all other segments, we found no tendency for TSS pe
13、aks (that is, in the first 200 nucleotides of a transcript) to be near a methylation consensus sequence (Supplementary Fig. 7b), suggesting that a methyltransferase other than the METTL3 complex is involved. Indeed, when adenosine is the first transcribed nucleotide, in addition totheobligatoryribos
14、e29-O-methylation,itcanbefurthermethylated byanothermethyltransferaseattheN6positionofthebasetogenerate (N6-,29-O-)-dimethyladenosine (m6Am)25,26, also recognizable by our antibody21. In support of this claim, transcripts with TSS peaks were ,25% more likely to begin with adenosine compared with tra
15、nscripts lackingsuchpeaks(SupplementaryFig.7a).Takentogether,TSSpeaks appeartoreflect,atleastpartly,m6Ambelongingtothe59capstructure, and attest, once more, to the validity of m6A-seq. We next examined the relationship between methylation and expression by plotting the fraction of genes with m6A pea
16、ks in each of the segments as a function of expression level. Interestingly, the CDS, stop codon and 39UTR segments exhibited a non-monotonic relationship: whereas transcripts of moderately expressed genes were morelikelytobemethylated,transcriptsofgenesexpressedatthetwo extremes were less methylate
17、d (Fig. 2d). This pattern, which is unlikely to be due to coverage limitations or to bias of differential read sampling (Supplementary Fig. 8), is of functional interest and is reminiscent of a similar relationship between DNA methylation found in gene bodies and their corresponding expression level
18、s27, raising the intriguing possibility that these phenomena might be con- nected. In contrast, we observed a positive correlation between expression leveland thepresence ofanm6Apeakinthe TSSsegment. Comparison of human and mouse methylomes The non-random distribution of m6A peaks in the human trans
19、crip- tome indicates a fundamental role of this modification. To determine the evolutionary conservation and consequent functional importance of m6A, the human and mouse methylomes were compared. We applied m6A-seq to RNA purified from mouse liver and obtained 4,513 m6A peaks within 3,376 coding gen
20、e transcripts and 66 non- coding ones (Supplementary Tables 1 and 7). Clustering of all signifi- cantly enriched sequences perfectly recapitulated the human methylation consensus sequence (Fig. 3a and Supplementary Fig. 9a). The mouse metagene profile revealed, as in human, a peak around the stop co
21、don and at the TSS (Fig. 3b and Supplementary Fig. 9c, d). In mouse,too,modificationswerehighlyenrichedinlonginternalexons, with 91% of the peaks in exons longer than 400 nucleotides (Supplementary Table 1 and Supplementary Fig. 9b). Thus, on the global level, m6A is highly conserved between the two
22、 species. We next systematically assessed the extent of m6A peak con- servation on the gene level. Of the 4,513 identified mouse m6A peaks, 2,023 could be reliably mapped to an adequately expressed, orthologous human position; 997 of them had an m6A peak in the orthologous human position, representi
23、ng 49% conservation. The highest extent of conservation is found at the stop codon segment, in which 57% of peaks are conserved, compared to 3249% in other segments. Conservation was also higher in internal exons longer than 400 nucleotides than in shorter ones (56% and 44%, respectively). The exten
24、t of conservation between human and mouse, which is highly significant over that which would be expected by chance (P53.73102136, MannWhitney test), is easily appreciated in the examplespresented (Fig.3candSupplementary Fig.10),andsignifies that these sites are likely to be functional. Methylation a
25、cross conditions and cells Accumulating evidence reveals that transfer RNA (tRNA) and ribo- somal RNA (rRNA) modifications change in response to stimuli, sug- gestingageneralmodelofdynamiccontroloverRNAmodification28,29. By analogy,we reasoned thatparticipation ofm6A inactive generegu- lationmay man
26、ifestitselfasaltered methylation profiles inresponseto changing cellular conditions or in a tissue-specific manner. We put our hypothesis to the test by comparing the methylation profiles of untreated HepG2 cells to those of cells exposed to ultra- violet radiation, heat shock, hepatocyte growth fac
27、tor (HGF; also known as scatter factor (SF), and interferon-c (Supplementary Table 2). Remarkably, all samples exhibited a marked similarity of Bits 2 1 0 2 1 m6A 1 2 3 3 4 a b 1,500 c Mean normalized coverage (a.u.) 5 Stop codon 400 400 400 TSS IP Input 3 1 Position (nt) 600 300 0 750 0 CDS start C
28、DS end IP Input 500 nt SLC39A14 No. of reads CDS start CDS end 500 nt Figure 3|m6A methylome conservation between human and mouse. a,Sequencelogorepresentingthededucedconsensusmotiffollowingclustering of all enriched motifs. b, Metagene profiles depicting sequence coverage in windows surrounding the
29、 TSS (left) and stop codon (right). Colour codes are the same as in panel Fig. 2a. The mouse stop codon segment is even more enriched in m6A peaks than its human counterpart, with 39% of all peaks located in this segment, corresponding to .3.5-fold enrichment over the distribution expected by chance
30、 (P,2.23102308, x2test). c, Orthologous genes with conserved m6A peaks. Gene architecture is shown beneath. a.u., arbitrary units. ARTICLE RESEARCH 1 0M A Y 2 0 1 2 |V O L4 8 5| N A T U R E|2 0 3 Macmillan Publishers Limited. All rights reserved2012 m6A profiles, with 7095% peak positions typically
31、shared between conditions (Supplementary Fig. 11). Nevertheless,wewereabletodetectasubsetoftreatment-dependent, dynamically altered peaks (Fig. 4 and Supplementary Table 8). These peaksdidnotcorrelatewithabsenceofaproximalconsensusmotif.We note that our stringent approach probably underestimates the
32、 amount ofdifferentialm6A peaks,all themoreasitisinsensitiveto the propor- tion of methylated versions of a specific transcript. We further examined tissue specificity and compared normal human brain to HepG2 cells. The former recapitulated all the global features of the m6A methylome initially iden
33、tified in HepG2 (includ- ingconsensus motifand peakdistribution along genearchitecture) to demonstrate that the characteristics of the methylome are not an aberration of cancer tissues but rather a normal phenomenon (Supplementary Note 2). m6A affects RNA splicing Thefunctionofm6AinRNAmetabolismwasn
34、extassessedbysilencing of METTL3 in HepG2 cells, as its depletion was already shown to reducetheamountofm6Ainthetranscriptome4.METTL3knockdown resulted in apoptosis (Supplementary Fig. 12), as expected3. Reads obtained from massively parallel sequencing of control (mock) and knockdown RNA were align
35、ed to the genome and differential gene expressionprofilesweregenerated(Methods;SupplementaryTable4). Of 1,977 differentially expressed genes, 1,218 contain mapped m6A peaks. Downregulated genes were significantly enriched with genes shown to have methylated introns (P51.331027, hypergeometric test;
36、Supplementary Fig. 13). We reasoned that focusing on genes that were not differentially expressed (fold change ,2) but whose constituent isoforms were (fold change .2), would help isolate a possible correlation between m6A and isoform switching (between up- and downregulated iso- forms). Indeed, met
37、hylated genes were overrepresented in this set (459/543, P57.9310212, hypergeometric test). Moreover, differ- entially spliced exons and introns were themselves significantly enriched with m6A peaks: 99/474 exons (P54310215, hypergeo- metric test) and 811/2,672 introns (P513102105, hypergeometric te
38、st), further supporting a role for m6A in splicing. ThesalientfindingsrevealedbyMETTL3knockdownpromptedus to examine all m6A peaks in the HepG2 transcriptome through the alternative splicing prism. Assigning peaks to either single- or multi- isoform coding genes revealed that the former are relative
39、ly under- methylated (6,489 and 11,946 peaks in 3,698 and 5,870 coding genes, respectively; P55.631027, hypergeometric test). Accordingly, the average number of m6A peaks per coding gene (1.93 peaks per gene) was higher in multi-isoform genes (2.04 peaks per gene) than in single-isoform ones (1.75 p
40、eaks per gene). Refining this observation further, assignment of all m6A peaks to either constitutive or alterna- tive spliced exons revealed that although 23.3% of coding exons are constitutive, they are significantly under-represented among all methylated sequences (10.9%, P,13102305, hypergeometr
41、ic test; Supplementary Fig. 14). Interestingly, GO analysis of differentially expressed genes indi- cated a noteworthy enrichment of the p53 signalling pathway (23 genes, corrected P56.031025): 22/23 genes had differentially expressed splice variants, of which 18 were methylated. Moreover, 15 other
42、members of the signalling pathway, which did not show significant differential expression at the gene level, exhibited signifi- cant differential expression at the isoform level (Supplementary Fig. 15a). For example, isoforms of MDMX (also known as MDM4), needed for p53 inactivation30, were downregu
43、lated (Supplementary Fig. 15b). Similar pro-apoptotic effects were observed in other key genesbelongingtothispathway(forexample,MDM2,FASandBAX). Modulation of p53 signalling through splicing may be relevant to induction of apoptosis by silencing of METTL3. KnockoutofIME4,theyeastorthologueofMETTL3,d
44、emonstrated its important role in regulation of the developmental switch from vegetative cells into gametogenesis, but failed to provide mechanistic insight (Supplementary Fig. 16). Binding of RNA-interacting proteins Methylation of RNA may also affect binding of interacting proteins, similartorecog
45、nitionof5-methylcytosineinDNAbyspecificbinding proteins that mediate its repressive effects31. We observed an overlap between known RNA sequence elements and newly identified m6A peaks: several internal ribosome entry sites (IRES)32, the localization zip code of b-actin33and the destabilization elem
46、ent of c-Myc (also known as MYC)34, all harbour methylation peaks (Supplementary Fig. 2b, c and Supplementary Fig. 17). An RNA affinity chromatography approach, using methylated and control versions of an RNA bait followed by mass spectrometry, was usedtoidentifynovelm6A-bindingproteins(Supplementar
47、yFig.18a). Our analysis identified three RNA-binding proteins that may mediate novel connections between m6A and cellular processes (Supplementary Fig. 18b and Supplementary Table 5). Two YTH (YT521-Bhomology)familyproteins, YTHDF2andYTHDF3,bound exclusively to the methylated bait (Supplementary Fig
48、. 18b, d). These proteins contain a recently characterized RNA-binding domain (YTH) that overlaps the methylated motif in our bait35(Supplemen- tary Fig. 18c). Interestingly, the only two characterized YTH family members, the human YT521-B (also known as YTHDC1) and the fission yeast Mmi1, were impl
49、icated in alternative splice site selection andin specifying transcripts fornuclear degradation, respectively36,37. 500 nt IFN- 0 1,000 UT 1,250 0 IP Input EGFR 0 750 0 750 No. of reads No. of reads PDE3A 500 nt UT CDS end IFN- CDS start Figure 4|Transcripts differing in methylation patterns under varying growing conditions. Arrows point at the differential position. The stability of the surrounding peaks underscores the authenticity of the changed peaks. Interferon-c (IFN-c) upregulates epidermal growth factor receptor (EGFR) RNA and protein levels47,48, suggesting an