LETTER doi:10.1038/nature12730 N6-methyladenosine-dependent regulation of messenger RNA stability Xiao Wang1, Zhike Lu1, Adrian Gomez1,GaryC.Hon2, Yanan Yue1, Dali Han1,YeFu1, Marc Parisien3, Qing Dai1, Guifang Jia1,4, Bing Ren2, Tao Pan3 & Chuan He1 N6-methyladenosine (m6A) is the most prevalent internal (non-cap) The YTH domain family is widespread in eukaryotes and known to modification present in the messenger RNA of all higher eukaryotes1,2. bind single-stranded RNA with the conserved YTH domain (.60% Although essential to cell viability and development3–5, the exact role identity) located at the C terminus16,17. In addition to previously reported of m6A modification remains to be determined. The recent discovery YTHDF2 and YTHDF314, we also discovered YTHDF1 as another m6A- of two m6A demethylases in mammalian cells highlighted the impor- selective binding protein by using methylated RNA bait containing the tance of m6A in basic biological functions and disease6–8.Herewe known consensus sites of G(m6A)C and A(m6A)C versus unmethy- show that m6A is selectively recognized by the human YTH domain lated control (Extended Data Fig. 1a). Further, highly purified poly(A)- family 2 (YTHDF2) ‘reader’ protein to regulate mRNA degradation. tailed RNAs were incubated with recombinant glutathione-S-transferase We identified over 3,000 cellular RNA targets of YTHDF2, most of (GST)-tagged YTHDF1-3 and then separated by GST-affinity column. which are mRNAs, but which also include non-coding RNAs, with a By using a previously reported liquid chromatography-tandem mass conserved core motif of G(m6A)C. We further establish the role of spectrometry (LC-MS/MS) method7,8, we found that the m6A-containing YTHDF2 in RNA metabolism, showing that binding of YTHDF2 RNAs were greatly enriched in the YTHDF-bound portion and dimin- results in the localization of bound mRNA from the translatable ished in the flow-through portion (Fig. 1b and Extended Data Fig. 1b). pool to mRNA decay sites, such as processing bodies9. The carboxy- Gel-shift assay revealed that YTHDF2 has a 16-fold higher binding terminal domain of YTHDF2 selectively binds to m6A-containing affinity to methylated probe compared to the unmethylated one, as mRNA, whereas the amino-terminal domain is responsible for the well as a slight preference to the consensus sequence (Extended Data localization of the YTHDF2–mRNA complex to cellular RNA decay Fig. 1c, d). This protein was selected for subsequent characterization 6 sites. Our results indicate that the dynamic m6A modification is because it has a high selectivity to m A, and was thought to be assoc- recognized by selectively binding proteins to affect the translation iated with human longevity18. status and lifetime of mRNA. Messenger RNA is central to the flow of genetic information. Regu- a b 1.65 latory elements (for example, AU-rich element, iron-responsive element), m6A methyltransferase 1.60 m6A binding protein in the form of short sequence or structural motif imprinted in mRNA, R R 1.20 HH A H CH N C N 3 are known to control the time and location of translation and degra- YTH N H N 10 N N CH3 0.80 dation processes . Reversible and dynamic methylation of mRNA (%) A/A N N N N 6 O O 0.42 m could add another layer of more sophisticated regulation to the prim- O mRNA O 0.40 0.21 2,11 6 O OH A O OH mRNA 6 ary sequence .m A, a prevalent internal modification in the messen- m A 0.00 ger RNA of all eukaryotes, is post-transcriptionally installed by m6A m6A demethylase Nucleus Cytoplasm methyltransferase (for example, MT-A70, Fig. 1a) within the consensus Input 6 6 12 sequence of G(m A)C (70%) or A(m A)C (30%) . The loss of MT-A70 Flow-through YTHDF2-bound 13 cdePeaks leads to apoptosis in human HeLa cells , and significantly impairs Non-coding YTHDF2 4 5 6 2.0 RNA development in Arabidopsis and in Drosophila . Our recent discoveries PAR-CLIP m A-seq Intron 15,455 ′ 1% 1% 6 7 12,442 3 UTR of m A demethylases FTO (fat mass and obesity-associated protein) 14% 8 Stop codon Intergenic and ALKBH5 demonstrate that this RNA methylation is reversible 1.0 3% 7,345 Bits 42% TSS 5,097 and may dynamically control mRNA metabolism. The recently revealed 3,512 11,943 5′ UTR 1% 6 CDS 2% m A transcriptomes (methylome) in human cells and mouse tissues 0.0 36% showed m6A enrichments within long exons and around stop codons14,15, 6 further suggesting fundamental regulatory roles of m A. However, Figure 1 | YTHDF2 selectively binds m6A-containing RNA. a, Illustration of 6 despite these progresses the exact function of m A remains to be m6A methyltransferase, demethylase and binding proteins. RRACH is the elucidated. extended m6A consensus motif, where R is G or A and H is not G. b, LC-MS/MS Whereas methyltransferase may serve as the ’writer’ and demethy- showing m6A enrichment in GST–YTHDF2-bound mRNA while depleted in lases (FTO and ALKBH5) act as the ‘eraser’ of m6A on mRNA, potential the flow-through portion. Error bars, mean 6 s.d., n 5 2, technical replicates. m6A-selective-binding proteins could represent the ‘reader’ of the m6A c, Overlap of peaks identified through YTHDF2-based PAR-CLIP and the m6A-seq peaks in the same cell line. d, Binding motif identified by MEME with modification and exert regulatory functions through selective recog- 246 PAR-CLIP peaks (P 5 3.0 3 10 , 381 sites were found under this motif out nition of methylated RNA. Here, we show that the YTH-domain family of top 1,000 scored peaks). e, Pie chart depicting the region distribution of member 2 (YTHDF2), initially found in pull-down experiments using YTHDF2-binding sites identified by PAR-CLIP, TSS (200-bp window from 6 6 m A-containing RNA probes14, selectively binds m A-methylated the transcription starting site), stop codon (400-bp window centred on mRNA and controls RNA decay in a methylation-dependent manner. stop codon). 1Department of Chemistry and Institute for Biophysical Dynamics, The University of Chicago, 929 East 57th Street, Chicago, Illinois 60637, USA. 2Ludwig Institute for Cancer Research, Department of Cellular and Molecular Medicine, UCSD Moores Cancer Center and Institute of Genome Medicine, University of California, San Diego School of Medicine, 9500 Gilman Drive, La Jolla, California 92093-0653, USA. 3Department of Biochemistry and Molecular Biology and Institute for Biophysical Dynamics, The University of Chicago, 929 East 57th Street, Chicago, Illinois 60637, USA. 4Department of Chemical Biology and Synthetic and Functional Biomolecules Center, College of Chemistry and Molecular Engineering, Peking University, Beijing 100871, China. 00 MONTH 2013 | VOL 000 | NATURE | 1 ©2013 Macmillan Publishers Limited. All rights reserved RESEARCH LETTER We next applied two independent methods to identify RNAs that ab 1.00 1.00 are the binding partners of YTHDF2: (1) photoactivatable ribounu- Non-targets (3,905) 19 CLIP targets (3,251) cleoside crosslinking and immunoprecipitation (PAR-CLIP) to locate CLIP+IP targets 0.75 0.75 the binding sites of YTHDF2; (2) sequencing profiling of the RNA (1,277) 20 of immunopurified ribonucleoprotein complex (RNP) (RIP-seq) 0.50 0.50 to extract cellular YTHDF2–RNA complexes. Approximately 10,000 crosslinked clusters covering 3,251 genes were identified in PAR-CLIP 0.25 0.25 (Extended Data Fig. 2a, b). Most are mRNA but 1% are non-coding Cumulative fraction P = 1.2 × 10–17 Cumulative fraction P = 0.28 P = 6.5 × 10–32 P = 1.5 × 10–7 RNA. Among 2,536 transcripts identified in RIP-seq, 50% overlap with 0.00 0.00 6 –1.5 0.0 1.5 –1.5 0.0 1.5 PAR-CLIP targets (Extended Data Fig. 2b). We also performed m A- log2(siYTHDF2/siControl) log2(siYTHDF2/siControl) seq for the poly(A)-tailed RNA from the same HeLa cell line and found Δ mRNA input Δ ribosome-protected fragments cd that 59% (7,345 out of 12,442) of the PAR-CLIP peaks of YTHDF2 1.00 t 1.00 I: Non-targets (2,400) 6 0 h < 1/2 < 24 h overlap with m A peaks (Fig. 1c). As shown in Fig. 1d, the conserved Non-targets (2,400) II: CLIP sites = 1 (1,034) III: CLIPsites = 2–4 6 CLIP targets (2,504) 0.75 (1,058) motif revealed from the top 1,000 scored clusters matches the m A 0.75 CLIP+IP targets 12,14 IV: CLIP sites = 5 consensus sequence of RRACH , which strongly supports the bind- (966) (412) ing of m6A by YTHDF2 inside cells (see more motifs in Extended Data 0.50 0.50 Fig. 2c–e). Coinciding with the previously reported pattern of m6A 0.25 0.25 14,15 P (I vs II) = 1.0 × 10–36 peaks , YTHDF2 PAR-CLIP peaks showed enrichment near the Cumulative fraction P = 1.7 × 10–103 P (II vs III) = 1.2 × 10–7 P = 1.4 × 10–139 P (III vs IV) = 1.1 × 10–8 stop codon and in long exons (Extended Data Fig. 2f–h). YTHDF2 0.00 0.00 predominantly targets the stop codon region, the 39 untranslated –1.5 0.0 1.5 –1.5 0.0 1.5 log2(siYTHDF2/siControl) log2(siYTHDF2/siControl) region (39 UTR), and the coding region (CDS) (Fig. 1e), indicating that Δ mRNA lifetime Δ mRNA lifetime YTHDF2 may have a role in mRNA stability and/or translation. e f siControl siYTHDF2 0.70 To dissect the role of YTHDF2 we used ribosome profiling to assess P = 0.002 P < 0.001 the ribosome loading of each mRNA represented as ribosome-protected 0.5 P < 0.001 0.60 0.54 21,22 0.4 0.48 0.49 reads .