Comprehensive Analysis of mRNA
Methylation Reveals Enrichment
in 30UTRs and near Stop Codons
Kate D. Meyer,1Yogesh Saletore,2,3Paul Zumbo,2,3Olivier Elemento,2,3Christopher E. Mason,2,3,* and Samie R. Jaffrey1,*
1Department of Pharmacology
2Department of Physiology and Biophysics
3HRH Prince Alwaleed Bin Talal Bin Abdulaziz Alsaud Institute for Computational Biomedicine
Weill Medical College, Cornell University, New York, NY 10065, USA
*Correspondence: email@example.com (C.E.M.), firstname.lastname@example.org (S.R.J.)
Methylation of the N6position of adenosine (m6A) is
understood prevalence and physiological relevance.
The recent discovery that FTO, an obesity risk gene,
encodes an m6A demethylase implicates m6A as an
important regulator of physiological processes.
Here, we present a method for transcriptome-wide
m6A localization, which combines m6A-specific
methylated RNA immunoprecipitation with next-
generation sequencing (MeRIP-Seq). We use this
method to identify mRNAs of 7,676 mammalian
genes that contain m6A, indicating that m6A is
a common base modification of mRNA. The m6A
modification exhibits tissue-specific regulation and
is markedly increased throughout brain develop-
ment. We find that m6A sites are enriched near stop
codons and in 30UTRs, and we uncover an associa-
tion between m6A residues and microRNA-binding
a resource for identifying transcripts that are
substrates for adenosine methylation and reveal
The fat mass and obesity-associated (FTO) gene is a major regu-
lator of metabolism and energy utilization (Church et al., 2009,
2010; Fischer et al., 2009). In humans, FTO polymorphisms
that increase FTO expression are associated with elevated
body mass index and increased risk for obesity (reviewed in
Fawcett and Barroso, 2010). FTO is a member of the Fe(II)-
and oxoglutarate-dependent AlkB oxygenase family and was
ylated thymidine and uracil (Gerken et al., 2007; Jia et al., 2008).
However, FTO exhibits low activity toward these base modifica-
tions, and they are relatively infrequent with unclear physiolog-
ical relevance (Klagsbrun, 1973). Thus, the physiologically
relevant targets of FTO were unclear until recent studies that
showed that FTO can demethylate N6-methyladenosine (m6A),
a naturally occurring adenosine modification (Jia et al., 2011).
These studies link adenosine methylation to physiological roles
in human biological processes.
The distribution of m6A in RNA is poorly understood. Previous
studies have found that m6A exists in RNA from a variety of
unique organisms, including viruses, yeast, and mammals (Bee-
mon and Keith, 1977; Bodi et al., 2010). m6A is found in tRNA
(Saneyoshi et al., 1969), rRNA (Iwanami and Brown, 1968), and
viral RNA (Beemon and Keith, 1977; Dimock and Stoltzfus,
1977). Although m6A is detectable in mRNA-enriched RNA frac-
tions (Desrosiers et al., 1974), it has been confirmed in vivo in
only one mammalian mRNA (Horowitz et al., 1984). The abun-
dance of m6A has been shown to be 0.1%–0.4% of total adeno-
1975; Wei et al., 1975), suggesting that this modification may
be widespread throughout the transcriptome. Although the
existence of m6A has been known for many years, progress in
establishing the prevalence of m6A in mRNA has lagged behind
that of other modified bases. This is due in large part to the lack
of available methods for detecting the presence of m6A.
Because methylation of adenosine does not alter its ability to
base pair with thymidine or uracil, m6A is not amenable to detec-
tion with standard hybridization or sequencing-based methods.
Here, we examine the prevalence, regulation, and functional
roles of m6A in the transcriptome. We show that m6A is a devel-
opmentally regulated RNA modification that is dynamically
modified. Using antibodies that recognize m6A, we have devel-
oped an affinity enrichment strategy that, when coupled with
next-generation sequencing, allows for the high-throughput
identification of m6A sites. Using this approach, we present the
first transcriptome-wide profile of m6A localization in RNA. We
find that m6A is a widespread modification that is present in
the mRNAs of >7,600 genes and in >300 noncoding RNAs.
Additionally, m6A is highly enriched near the stop codon and in
the 30UTR. Furthermore, bioinformatic analysis of m6A localiza-
tion reveals consensus sites for m6A and identifies a potential
interaction between m6A and microRNA pathways.
Cell 149, 1635–1646, June 22, 2012 ª2012 Elsevier Inc. 1635
Detection of m6A in Mammalian mRNA
Because m6A exhibits the same base pairing as unmodified
adenosine, it is not readily detectable by standard sequencing
or hybridization-based approaches. Additionally, m6A is not
susceptible to chemical modifications that might otherwise facil-
itate its detection, such as bisulfite treatment, which is used to
detect 5mC in DNA. The methods used thus far to detect m6A
have involved treating cells with radiolabeled methionine, the
precursor of the endogenous methylating agent S-adenosylme-
thionine, to impart radiolabeled methyl groups to adenosine
(Csepany et al., 1990; Dubin and Taylor, 1975; Narayan and
Rottman, 1988). Radiolabeled m6A residues are subsequently
mapped with thin-layer chromatography or HPLC.
In order to simplify detection of m6A, we sought to develop an
immunoblotting strategy. For these experiments, we used
a previously described anti-m6A antibody (Bringmann and Lu ¨hr-
mann, 1987; Jia et al., 2011; Munns et al., 1977). To ensure the
specificity of thisantibody form6A,weperformed dot blots using
modified oligonucleotides immobilized to a membrane. The m6A
antibody selectively bound to oligonucleotides containing a
single m6A residue and exhibited negligible binding to oligonu-
cleotides containing unmodified adenosine (Figure 1A). The
binding was competed by incubating the antibody with
increasing concentrations of an m6A-rich competitor RNA (Fig-
ure 1B). However, RNA containing unmodified adenosine did
not compete for binding. Furthermore, binding was competed
by N6-methyladenosine triphosphate, but not by ATP or other
modified adenosine triphosphates, including N1-methyladeno-
sine and 20-O-methyladenosine (Figure 1C). Finally, to examine
the specificity of the antibody in the context of other nucleotide
sequences, we took advantage of the fact that the enzyme
encoded by the DNA adenine methylase (dam) gene in E. coli
digested DNA isolated from dam+ and dam- E. coli to immuno-
blotting using the m6A antibody, we found robust signals only in
the DNA samples from the dam+ strain (Figure 1D). Collectively,
these data demonstrate the high sensitivity and selectivity of this
antibody for m6A, as well as its ability to detect m6A within
cellular nucleotide pools.
To explore the abundance of m6A within various RNA
populations, we isolated RNA from several mouse tissues and
subjected it to immunoblot analysis using the m6A antibody (Fig-
ure 2A). We found that m6A was present in all RNA samples
tested, indicating that this modified nucleotide is widely distrib-
uted in many tissues, with particularly high enrichment in liver,
kidney, and brain (Figure 2B). In addition, we observed large
differences in the m6A content of various immortalized cell lines,
including several cancer cell lines, which further indicates that
large differences in m6A levels exist in different cell populations
(Figure S1A available online).
m6A immunoreactivity was detected in bands throughout the
with incorporation of m6A in mRNA. Indeed, fractionation of
whole cellular RNA into polyadenylated and nonpolyadenylated
Figure 1. Specificity and Sensitivity of m6A-
(A) Dot blot analysis demonstrates antibody specificity for
m6A. Increasing amounts of an oligonucleotide containing
either m6A or unmodified adenosine were spotted onto
a membrane and probed with the m6A antibody. Though
increased m6A immunoreactivity is observed in the pres-
ence of increasing concentrations of the m6A oligonucle-
otide(top),onlybackground levelsofimmunoreactivity are
observed at the highest concentrations of the A oligonu-
cleotide (bottom). Blots shown are representative of
results from three experiments.
(B) Competition dot blot assays were performed on
membranes spotted with 100 ng of m6A-containing
oligonucleotide. Antibody binding to the m6A oligonucle-
otide is attenuated by preincubation with increasing
amounts of m6A-containing competitor RNA (top), but not
with RNA containing unmodified adenosine (bottom).
Amount of competitor RNA used (left to right): 0 ng (0 nM),
10 ng (0.1 nM), 100 ng (1.1 nM), 1 mg (11.2 nM). Blots
(C) Competition dot blot assays were performed as in (B).
Antibody was preincubated with increasing amounts
of N6-methyladenosine triphosphate (N6-MeATP), aden-
osine triphosphate (ATP), N1-methyladenosine triphos-
phate (N1-MeATP), or 20-O-methyladenosine triphosphate
(20-O-MeATP). Only N6-MeATP is able to compete with
antibody binding. Concentration of competitor nucleotide
used (left to right): 0 mM, 1 mM, 2 mM, 4 mM. Blots shown are representative of results from three experiments.
(D) Detection of m6A in cellular DNA. Genomic DNA isolated from dam+ (containing m6A) or dam? (lacking m6A) E. coli was sheared and subjected to
immunoblotting with the m6A antibody. Although 1.5 times as much DNA from dam? E. coli was loaded (left), the antibody only recognizes the m6A present in
DNA from dam+ E. coli (right). Blot shown is representative of results from three experiments.
See also Figure S2.
1636 Cell 149, 1635–1646, June 22, 2012 ª2012 Elsevier Inc.
RNAs indicates that m6A immunoreactivity is enriched in the pol-
yadenylated RNA pool, which further suggests that m6A in
cellular RNA is localized to mature mRNA (Figure 2C). To deter-
mine whether m6A is present in poly(A) tails, we selectively
removed the poly(A) tail from cellular mRNA using oligo(dT)
hybridization and RNase H treatment. Transcripts depleted of
the poly(A) tail did not exhibit an appreciable reduction in m6A
levels (Figure 2D). In addition, immunoblotting poly(A) tails alone
showed minimal m6A immunoreactivity (Figures S2C–S2E).
Together, these data demonstrate that m6A is primarily an
internal modification that is largely absent from the poly(A) tail.
Dynamic Regulation of m6A
Our observation that m6A is highly enriched within the brain
prompted us to investigate the temporal dynamics of m6A levels
during different stages of neural development. Immunoblotting
in mRNA at low levels throughout embryogenesis but increases
dramatically by adulthood (Figure 3A). A similar increase in m6A
upregulation of m6A levels accompanies neuronal maturation.
We next asked whether adenosine methylation is a dynami-
cally regulated posttranscriptional modification and whether its
levels can be regulated by specific demethylating enzymes. In
our search for potential demethylating enzymes that act to re-
move the methyl group from m6A, we focused on members of
the family of Fe(II)- and 2-oxoglutarate-dependent oxygenases,
several of which have previously been shown to demethylate
both DNA and RNA (Falnes et al., 2007; Gerken et al., 2007).
Consistent with the findings of Jia et al. (2011), we observed
that FTO decreased m6A levels when overexpressed in mamma-
FTO resulted in a broad size range of RNAs that exhibit reduced
m6A immunoreactivity (Figure 3B).
MeRIP-Seq Identifies m6A-Containing RNAs throughout
In order to obtain insight into potential roles for m6A, we sought
to characterize its distribution throughout the transcriptome. To
do this, we first determined whether the m6A antibody could be
used to enrich m6A-containing RNAs. In vitro immunoprecipita-
tion experiments showed that a single round of MeRIP produces
?70-fold enrichment, and two roundsproduce >130-fold enrich-
ment for m6A-containing targets (Figure S3). To identify m6A
sites throughout the transcriptome, we developed a method
that combines m6A-specific methylated RNA immunoprecipita-
tion (MeRIP) with next-generation sequencing (RNA-Seq). The
procedure for MeRIP-Seq (outlined in Figure 4A) involves
randomly fragmenting the RNA to ?100 nt sized fragments prior
to immunoprecipitation. Because an m6A site could lie anywhere
along the length of a given immunoprecipitated 100 nt fragment,
sequencing reads are expected to map to a region that contains
the m6A site near its center. At its extremes, this region would be
predicted to be roughly 200 nt wide (100 nt up- and downstream
from the m6A site) (Figures 4B and 4C).
Figure 2. Distribution and Dynamic Cellular Regu-
lation of m6A in RNA
(A) Widespread distribution of m6A levels in a variety of
tissues. Total RNA isolated from mouse brain, heart, lung,
liver, and kidney (top) was subjected to m6A immunoblot
analysis. Ethidium bromide staining of the 28S rRNA is
shown as a loading control (bottom).
(B) Quantification of m6A abundance within various
tissues. Quantification of m6A immunoreactivity in (A) was
measured by densitometry and normalized to the intensity
data are presented as mean ± SEM).
(C) m6A is enriched within mRNAs. Oligo(dT) Dynabeads
were used to isolate poly(A) RNA from total mouse brain
RNA, and the unbound ‘‘flowthrough’’ RNA was saved as
the poly(A)-depleted fraction. Equal amounts of total RNA,
poly(A) RNA, and poly(A)-depleted RNA were then
subjected to m6A immunoblot analysis (top). Ethidium
bromide staining of 28S rRNA is shown as a loading
observed in the poly(A) RNA fraction, consistent with high
levels of m6A within mRNAs.
(D) Depletion of poly(A) tails from mRNA does not reduce
levels of m6A in mRNA. Poly(A) RNA was isolated from
total mouse brain RNA using oligo(dT) Dynabeads. Half
of the sample was then subjected to poly(A) tail depletion
by hybridizing to oligo(dT) primers and digestion with
shows that levels of m6A in poly(A) RNA (left) and poly(A)
tail-depleted RNA (right) are comparable. Removal of
poly(A) tails was confirmed using 30RACE and RT-PCR to detect b-actin; no product is detected in the tail-depleted sample when oligo(dT) primers are used for
cDNA synthesis (middle). As a control, use of random hexamers successfully generates a product in both samples (bottom).
See also Figure S1.
Cell 149, 1635–1646, June 22, 2012 ª2012 Elsevier Inc. 1637
We next utilized MeRIP-Seq to identify m6A sites in total
mouse brain RNA. Reads from the MeRIP sample frequently
mapped to mRNAs and clustered as distinct peaks. As pre-
dicted, these peaks frequently converged to ?100 nt wide
regions near their midpoint (Figure 4C). Furthermore, enrichment
of reads in these regions was not observed in the non-IP control
sample, which was composed of the input RNA prior to m6A
immunoprecipitation, demonstrating the specificity of these
peaks (Figure S3).
To determine the location of these peaks throughout the
transcriptome and thus characterize the regions of m6A localiza-
tion, we developed an algorithm for identifying m6A peaks
(see Extended Experimental Procedures). Additionally, we per-
formed replicate MeRIP-Seq experiments in which we utilized:
(1) a different sequencing platform (Illumina’s GAIIx versus
HiSeq2000), (2) independently prepared RNA samples from
different animals, and (3) an unrelated m6A antibody (Kong
et al., 2000), which exhibited similarly high specificity for m6A
(Figure S4). We employed our algorithm to identify m6A peaks
that met a minimum p value (p % 0.05, Benjamini and Hochberg
corrected) within each individual sample. From the three
samples, we identified a total of 41,072 distinct peaks in the
RNAs of 8,843 genes, which we call our ‘‘filtered’’ set of m6A
peaks (Table S1). Of these peaks, 80% were detected in at least
two different replicates. The high concordance between these
samples indicates that MeRIP-Seq is highly reproducible across
different sequencing platforms and using different m6A anti-
bodies. For subsequent bioinformatic analyses, we used the
list of 13,471 m6A peaks in RNAs from 4,654 genes that were
detected in all three replicates (our ‘‘high-confidence’’ list; Table
S2A). This list demonstrates the presence of m6A in a substantial
feature of mammalian mRNAs.
m6A Is Detected in Noncoding RNAs
The majority of our high-confidence m6A peaks (94.5%) are
found within mRNAs. However, we also observed that 236
(1.8%) of our peaks mapped to noncoding RNAs (ncRNAs) that
were annotated in the RefSeq database (Table S2A). In addition,
To determine whether these unannotated peaks localize to
ncRNAs predicted in other databases, we aligned them to
genomic regions of a set of 32,211 ncRNAs from the RIKEN
functional annotation of mouse (FANTOM3) data set that we
obtained from the mammalian noncoding RNA database
(RNAdb; Pang et al., 2005). We found that 216 of these peaks
mapped to a FANTOM3 ncRNA (Table S2B). All of these ncRNAs
were >200 nt in length, indicating that long ncRNAs are
substrates for adenosine methylation. Additionally, when we
interrogated a set of conserved human lincRNAs (Cabili et al.,
2011) for overlaps with m6A peaks, we found nine additional
peaks that overlapped with these lincRNAs (Table S2C). Collec-
tively,thesedataidentifyseveral classes ofncRNAsastargetsof
Biochemical Validation of m6A-Containing Transcripts
We next sought to validate the presence of m6A in mRNAs
identified with MeRIP-Seq. To do this, we used RNA pull-down
assays to isolate individual mRNAs from total mouse brain
RNA by hybridization to target-specific probes. Isolated mRNAs
were then subjected to immunoblot analysis using the m6A
antibody to detect the presence of m6A. Using this method, we
validated the presence of m6A within low density lipoprotein
receptor (Ldlr) (Figures 5A and 5B), metabotropic glutamate
receptor 1 (Grm1), and dopamine receptor D1A (Drd1a) (Figures
S5A–S5D). These mRNAs were chosen to demonstrate our
ability to validate m6A presence in transcripts with multiple
methylation sites (Grm1 and Drd1a) as well as those with single
m6A peaks (Ldlr). To further demonstrate that MeRIP-Seq selec-
tively enriches for these endogenous methylated targets, we
performed qRT-PCR on the unbound fractions after RNA precip-
itation with the m6A antibody. As expected, we observed
substantial immunodepletion of Grm1, Drd1a, and other methyl-
ated targets in the unbound fraction. In contrast, transcripts that
lack m6A peaks, such as Rps21 and Ndel1, were detectable at
high levels in the unbound fraction (Figure S5E).
m6A-Containing mRNAs Are Involved in Important
To predict potential signaling pathways and cellular processes
that involve m6A, we used the DAVID bioinformatics database
to identify the gene ontology (GO) terms that are enriched for
m6A-containing transcripts. We found that genes encoding
m6A-containing RNAs are involved in a variety of cellular func-
tions, including transcriptional regulation, RNA metabolism,
and intracellular signaling cascades (Table S5). In addition,
we observed that m6A peaks mapped to many genes linked to
neurodevelopmental and neurological disorders, such as Bdnf,
Dscam, Lis1, and Ube3a, as well as the neurexins and several
neuroligins. Collectively, these data demonstrate that m6A-
containing RNAs are involved in a variety of biological pathways
that are relevant to cellular signaling and disease.
Because m6A is a physiological target of FTO, we sought
to determine whether mRNAs whose levels have previously
been shown to be influenced by FTO activity contain m6A. We
Figure 3. Regulation of m6A Levels in Cells and during Development
(A) Ontogeny of m6A abundance throughout brain development. Total RNA
postnatal day 14 (P14), and adulthood and was then subjected to immunoblot
analysis to detect m6A-containing transcripts. Ethidium bromide staining of
28S rRNA bands is shown as a loading control.
(B) FTO demethylates a wide range of cellular transcripts. FTO was expressed
in HEK293T cells for 48 hr, and cellular RNA was subjected to immunoblot
analysis to detect m6A.
See also Figure S1.
1638 Cell 149, 1635–1646, June 22, 2012 ª2012 Elsevier Inc.
examined a list of 77 mRNAs whose levels are either up- or
downregulated in the liver, skeletal muscle, or white adipose
tissue of mice homozygous for a nonsynonymous FTO point
mutation (Church et al., 2009). mRNAs from seven genes that
were significantly upregulated in FTO mutants (Acaca, Atf6,
Bip, Gcdh, Irs1, Perk, and Xbp1) also contain m6A peaks. Intrigu-
ingly, some of these genes are involved in important metabolic
pathways, raising the possibility that demethylation of the tran-
scripts of these genes may contribute to the mechanism by
which FTO regulates metabolism and energy homeostasis.
Diverse Patterns of m6A Localization within Transcripts
We next characterized the pattern of adenosine methylation in
mRNAs. mRNAs from many genes (46.0%) exhibit a single m6A
peak, consistent with a single m6A site or a cluster of adjacent
m6A residues. However, 37.3% contain two m6A peaks, 11.2%
contain three peaks, and 5.5% contain four or more peaks.
genes that exhibited the largest number of m6A peaks, all had 15
or more m6A peaks along their length (Table S3). Additionally, of
the genes that contain more than one m6A peak, 90.1% contain
two or more contiguous m6A peaks, suggesting that m6A sites
are frequently clustered in adjacent regions along the transcript.
or more adjacent m6A peaks, suggesting that m6A clustering is
a common feature in methylated transcripts (Figure S6A). We
multiple m6A residues throughout these regions.
We next determined which mRNAs contain m6A sites with the
highest degree of methylation. To do this, we developed
a method of calculating the level of m6A enrichment at individual
m6A peaks, which normalized the number of MeRIP sample
reads within each peak to the abundance of the individual tran-
script in which the peak resides (see Extended Experimental
Procedures). The genes that contain the most enriched m6A
peaks are listed in Table 1. Importantly, because MeRIP-Seq
identifies m6A sites at a resolution of 200 nt, there could be
multiple individual m6A residues within the area covered by
each peak. Therefore, the peaks with the highest levels of local
m6A enrichment may represent a single adenosine residue that
exhibits a high degree of methylation or multiple adjacent m6A
residues with a lower stoichiometry of methylation. In either
case, the high levels of methylation observed at these sites likely
indicate transcripts that are most influenced by m6A-dependent
m6A Is Enriched near Stop Codons and in 30UTRs
the transcriptome in our high confidence set. The majority
These m6A peaks are abundant in coding sequences (CDS;
50.9%) and untranslated regions (UTRs; 41.9%), with relatively
few in intronic regions (2.0%) (Figure 5C). Additionally, m6A
peaks are less abundant in the 50UTR (7.0% of UTR peaks)
than in the 30UTR (93.0% of UTR peaks) (Figure 5C). This distri-
bution deviates substantially from the distribution of reads in
the non-IP sample, indicating the high degree of enrichment of
Figure 4. Outline of MeRIP-Seq Protocol
and Distribution of Sequencing Reads
(A) Schematic representation of MeRIP-Seq. Total
RNA is subjected to RiboMinus treatment to
remove rRNA species. RNAs containing m6A are
then immunoprecipitated by mixing the RNA with
m6A antibody-coupled Dynabeads. m6A-contain-
ing RNAs are then eluted from the antibody-
coupled beads and subjected to a second round
of m6A immunoprecipitation. The resulting RNA
pool, which is highly enriched for m6A-containing
RNAs, is then subjected to next-generation
(B) Schematic of sequencing reads and their
alignment to locations in the genome surrounding
an m6A site. (Top) An mRNA that contains a single
m6A residue along its length. (Middle) Individual
100 nt wide mRNA fragments that are isolated
following m6A immunoprecipitation, each of which
contains the same m6A residue from the mRNA
depicted above. (Bottom) Histogram showing
predicted frequency of MeRIP-Seq reads ob-
tained by sequencing individual immunoprecipi-
tated fragments. Read frequency is predicted to
increase with closer proximity to the m6A site,
forming a ‘‘peak’’ that is roughly 200 nt wide at its
base and 100 nt wide at its midpoint.
(C) Sequencing reads from MeRIP-Seq converge
of m6A (shown here is the 30UTR of Pax6). Peak height is displayed as reads per base per million mapped reads (BPM).
See also Figures S2 and S3.
Cell 149, 1635–1646, June 22, 2012 ª2012 Elsevier Inc. 1639
Figure 5. Validation of m6A Targets and Characteristics of m6A Localization
(A) Different sequencing platforms and antibodies result in similar m6A profiles. UCSC Genome Browser tracks displaying read clusters from three MeRIP-Seq
replicates (MeRIP1, MeRIP2, and MeRIP3) are shown along the length of the Ldlr transcript. The upper-most track (non-IP) represents the non-
immunoprecipitated control sample.
(B) Validation of m6A-containing mRNA identified with MeRIP-Seq. Hybridization-based RNA pull-down was used to isolate Ldlr mRNA from total brain RNA,
followed by confirmation of m6A presence (arrow) by immunoblot analysis with anti-m6A. A control sample using a nonspecific probe of equal size (CTL Probe)
was run in parallel. Total mouse brain RNA (Input) is shown as a reference for m6A labeling.
(C) Transcriptome-wide distribution of m6A peaks. Pie charts showing the percentage of m6A peaks (top) and non-IP sample reads (bottom) within distinct RNA
sequence types. m6A is highly enriched in 30UTRs and CDSs compared to the distribution of reads in the non-IP, rRNA-depleted samples.
(D) Distribution of m6A peaks across the length of mRNA transcripts. 50UTRs, CDSs, and 30UTRs of RefSeq mRNAs were individually binned into regions
spanning 1% of their total length, and the percentage of m6A peaks that fall within each bin was determined. The moving averages of mouse brain peak
percentage (red) and HEK293T peak percentage (blue) are shown.
(E) Highly similar m6A peak distribution is observed within many human and mouse transcripts. UCSC Genome Browser plots showing MeRIP-Seq read clusters
in the representative transcript SREK1. MeRIP-Seq reads cluster at the same distinct regions of SREK1 in both HEK293T cell RNA (top) and mouse brain RNA
See also Figures S4–S7 and Tables S1–S6.
1640 Cell 149, 1635–1646, June 22, 2012 ª2012 Elsevier Inc.
m6A peaks in the CDS and UTRs (Figure 5C). Although a
low percentage of m6A peaks was observed in intronic regions,
because our samples were not enriched for unspliced pre-
mRNAs, it is possible that additional methylated intronic
tially found in certain portions of transcripts. To do this, we
assigned each m6Apeak to either a50UTR, CDS,or 30UTRcate-
gory and assigned it to 1 of 100 bins based on its location along
the 50UTR, CDS, or 30UTR. These data show that m6A occurs
at low levels in the 50UTR and in the 50end of the CDS. In the
script length and is, on average, 5- to 6-fold higher at the end of
the CDS than at the beginning (Figure 5D). In the 30UTR, the
peaks are enriched near the stop codon and decrease in
abundance along the length of the 30UTR. Indeed, 61% of
m6A peaks are in the first quarter of the 30UTR, and a quarter
of all m6A peaks across the entire transcriptome are found within
the first 26% of the 30UTR (Figure 5D). Mapping the number of
m6A peaks 1 kb up- and downstream of CDS end sites further
demonstrated the high levels of methylation in the vicinity of
the stop codon (Figures S7A and S7B). Collectively, these data
indicate that m6A peaks are highly clustered in the vicinity of
the stop codon in mRNAs.
m6A Occurs in Highly Conserved Regions within Unique
Wenextaskedwhetherm6Asitesareconserved across species.
We compared PhyloP scores across 30 vertebrates (Pollard
et al., 2010) of m6A peak regions to those of random regions of
the same size in gene exons. We found that the distribution of
conservation scores of the m6A peaks was significantly different
from that of the random regions (p % 2.2 3 10?16, Kolmogorov-
Smirnov [K-S] test; Figure 6A) and that m6A peaks’ median
conservation score (0.578) was much higher than that of the
random regions (0.023). The fact that m6A frequently occurs in
evolutionarily conserved sequences suggests that m6A-contain-
ing regions are maintained through selection pressure.
Because the tools for transcriptome-wide localization of m6A
sites have until now been unavailable, only a few studies to
date have examined the sequence contexts of m6A formation
(Canaani et al., 1979; Dimock and Stoltzfus, 1977; Wei et al.,
1976). Using methods such as RNase T1 fingerprinting of radio-
labeled RNA followed by separation by thin-layer chromatog-
raphy, these studies reported that m6A exists within two unique
sequence contexts: GAC and AAC (underlined adenosines
indicate m6A). Subsequently, an extended m6A consensus
sequence was identified: PuPuACX (Pu = purine; X = A, C, or
U). However, because the methods used in these studies are
not practical for use in a high-throughput manner, it is unclear
whether these motifs are relevant to the transcriptome-wide
m6A sites identified by MeRIP-Seq.
We therefore sought to identify sequence motifs that are
enriched within m6A peaks. To do this, we used FIRE, a sensitive
and unbiased tool for discovering RNA regulatory elements
(Elemento et al., 2007). Remarkably, FIRE independently identi-
fied the GAC and AAC motif, G[AG]ACU, and related variants
([AC]GAC[GU], GGAC, [AU][CG]G[AG]AC, and UGAC) as being
Table 1. Genes Encoding Transcripts with the Highest Degree of m6A Enrichment
Chr Peak StartPeak End RefSeq AccessionGene Symbol Enrichment Score
chr6 58856032 58856214 NM_021432Nap1l5 3.859
chr330717625 30717825NM_027016Sec62 3.693
chr3 8856606788566234NM_018804 Syt113.452
chr195801367 5801534NR_002847Malat1 3.322
chr12 110898950110899150 NR_028261Rian3.222
chr10 3401883434018993 NM_030203Tspyl43.162
chr176138575 6138775 NM_054040Tulp42.984
chr234631357 34631514 NM_001163434Hspa52.935
chr2 102630175102630350 NM_001077514 Slc1a22.933
chr117072225 7072425NM_009622 Adcy12.865
chr2158211200158211350 NM_175692 Snhg112.788
chr1537326800 37327000NM_134094 Ncald2.685
chr2 102629600102629788 NM_001077514 Slc1a2 2.672
chr15 3732700037327200NM_134094 Ncald2.656
chr10 8064495080645125 NM_007907 Eef22.451
chr2102629788102629975 NM_001077514Slc1a2 2.392
chr11 5925161359251750 NM_144521Snap47 2.306
chr1574581075 74581267NM_011838 Lynx1 2.258
Shown are the top 20 genes that contain m6A peaks with the highest levels of enrichment.
158211050 158211200NM_175692 Snhg112.102
Cell 149, 1635–1646, June 22, 2012 ª2012 Elsevier Inc. 1641
Figure 6. MeRIP-Seq Reveals Features of m6A in mRNA
(A) Phylogenetic conservation of m6A peaks. PhyloP scores of m6A peak regions were compared to those of randomly shuffled regions throughout gene exons.
There was a significantly higher median conservation score (K-S test, *p % 2.2 3 10-16) in m6A peaks (0.578) than in the random regions (0.023).
(B) Sequence motifs identified within m6A peaks. The motif G[AG]ACU and variants thereof ([AC]GAC[GU], GGAC, [AU][CG]G[AG]AC, and UGAC) were highly
motif indicate the degree of underrepresentation (blue) or overrepresentation (yellow) within regions of m6A peaks in the non-IP control sample (CNTL) and the
MeRIP sample (MeRIP).
(C) m6A motif sequences frequently lie near the center of m6A peaks. Shown is a plot of the cumulative distribution of m6A motif positions within m6A peaks
containing a single motif. Motifs cluster in the center of peaks, suggesting that the methylated adenosines in these motifs account for the m6A peaks identified in
(D) Example of a m6A motif sequence near the center of a peak. UCSC Genome Browser plot containing tracks for MeRIP-Seq reads (red) and non-IP control
reads (black) at the Ilf2 locus. The m6A peak within the Ilf2 30UTR contains a single m6A motif identified in (B). The sequence of this motif (highlighted in yellow) is
located at the center of the m6A peak.
(E) Distribution of m6A peaks and miRNA target sites within 30UTRs. The frequencyof m6A peaks (blue) and miRNA target sites(red)along thelength of 30UTRs is
(F) Associationbetween 30UTR methylationand miRNAabundance.The25most abundant miRNAsinbrain haveasignificantlygreater percentageofm6Apeaks
within their target mRNA 30UTRs than do the 25 most weakly expressed brain miRNAs (*p < 0.05, Wilcoxon test).
The error bars in (A) and (F) indicate the highest and lowest values, and the box boundaries denote the first quartile, median, and third quartile.
See also Figures S3, S6, and S7.
1642 Cell 149, 1635–1646, June 22, 2012 ª2012 Elsevier Inc.
highly enriched in m6A peaks (Figure 6B). For example, the
G[AG]ACU motif occurs in 42% of all m6A mRNA peaks and in
a much lower fraction (21%) of non-m6A control peaks from
the same mRNAs (p < 1.0 3 10?124, chi-square test). Altogether,
we found that >90% of all m6A peaks contain at least one of the
motifs identified by FIRE.
Wenext examined the position ofthe motifs within m6Apeaks.
Nearly 30% of m6A peaks have only one motif (Figure S6B), indi-
cating that these peaks are likely to contain only a single methyl-
ated residue. Motifs are also preferentially found in the center of
m6A peaks (Figures 6C and 6D), suggesting that these peaks
derive from a centrally located methylated adenosine residue.
Of note, other RNA regulatory elements, such as AU-rich
elements, poly(A) signals, or binding sites for known RNA-
binding proteins, were not identified by FIRE, suggesting that
m6A is unlikely to primarily function by modifying these known
Relationship between m6A Sites and Polyadenylation
Signals in 30UTRs
FIRE did not identify an enrichment of poly(A) signals (PASs),
which are involved in 30UTR end processing, in m6A peaks.
However, PASs exhibit considerable sequence heterogeneity
beyond the canonical AAUAAA consensus (Tian et al., 2005).
This sequence heterogeneity might allow these PASs to evade
detection by FIRE despite being enriched in m6A peaks.
Therefore, we sought to further investigate whether m6A peaks
within 30UTRs are enriched at PASs. We obtained a high-confi-
dence list (Brockman et al., 2005) of poly(A) cleavage sites (the
site downstream of a PAS where the mRNA is actually cleaved
and polyadenylated) for the mRNAs that contain m6A peaks
within their 30UTRs. We then examined whether m6A peaks
were enriched near these sites by determining the number of
30UTR m6A peaks that fell within 50 nt upstream of each
cleavage site. Because a PAS is located ?10–30 nt upstream
of an actual mRNA cleavage site (reviewed in Proudfoot, 1991),
these 50 nt long regions are expected to contain the PAS. Of
the 6,288 m6A peaks found within 30UTRs, 1,042 (16.6%)
overlapped with the 50 nt long regions upstream of poly(A)
cleavage sites, compared to 1,070 (17.0%) control peaks, which
were generated from random nonoverlapping regions of the
same 30UTRs. Thus, these data suggest that m6A does not
have a significant association with known PASs (p = 0.39;
m6A Is Not Enriched at Splice Junctions
Prior studies that used nonspecific methylation inhibitors to
explore possible functions for m6A revealed impaired splicing
in a small number of RNAs (Carroll et al., 1990; Stoltzfus and
Dane, 1982). We therefore asked whether the localization
of m6A peaks is compatible with a role for influencing the binding
of splicing factors. However, only 80 splice junctions were found
in regions contiguous with m6A peaks, significantly fewer
than the overlap seen with a set of randomly generated peaks
(9,531; p = 0.0; chi-square test). Thus, unlike CLIP-Seq tag
clusters from RNA-binding proteins that influence splicing
(Licatalosi et al., 2008), m6A peaks did not significantly coincide
with exon-exon junctions, suggesting that m6A is unlikely
to primarily function to directly influence the binding of splicing
Relationship between m6A and MicroRNA-Binding Sites
The strong enrichment of m6A peaks in 30UTRs prompted
us to investigate whether m6A peaks are found near microRNA
(miRNA)-binding sites, which are also frequently observed within
30UTRs. We found that 67% of 30UTRs that contain m6A peaks
also contain at least one TargetScan-predicted miRNA-binding
site. Because ?30% of genes have miRNA-binding sites in
their 30UTRs (Lewis et al., 2005), this is a significantly greater
association than what would be expected by chance alone.
Intriguingly, we also found that, in 30UTRs with both m6A peaks
and miRNA-binding sites, the m6A peaks precede miRNA-
binding sites 62% of the time. Moreover, we found that the
overall distribution of m6A peaks and miRNA-binding sites within
30UTRs is anti-correlated; whereas m6A peaks are most
abundant near the stop codon and generally decrease in
frequency along 30UTR length, miRNA target sites are more
enriched near the 30end of 30UTRs (Figure 6E). The reason for
this inverse localization pattern is unknown, though it could
indicate that a certain spatial separation is necessary for m6A
to influence the function of a downstream-bound miRNA or
We next sought to determine whether miRNA-targeted tran-
scripts in the brain are more likely to contain m6A. To test this,
we used TargetScan to identify the target transcripts of the 25
most highly expressed and 25 least highly expressed miRNAs
within the brain. Intriguingly, we observed that the most highly
expressed miRNAs have a significantly greater percentage of
target transcripts that contain m6A (p < 0.05, Wilcoxon test; Fig-
ylation of their target transcripts.
Prominent Features of m6A Distribution Are Conserved
in the Human Transcriptome
We next asked whether the enrichment of m6A in the 30UTR
is also observed in other species. We therefore profiled m6A in
HEK293T cells, a human cell line with high levels of adenosine
methylation (Figure 3B). We generated a high-confidence list
of m6A peaks using three MeRIP-Seq biological replicates and
confirmed by both m6A antibodies. We found that the distribu-
tion of m6A peaks in HEK293T cells closely mirrored the distribu-
tion in mouse brain, with 31% and 53% of m6A peaks falling
within the 30UTR and the CDS, respectively (Figures 5D and
S7D). As with the pattern of m6A distribution in the mouse brain
transcriptome, HEK293T m6A peaks were predominantly local-
ized near stop codons (Figures 5D and S7C).
In total, we identified 18,756 peaks in RNAs encoded by 5,768
genes in HEK293T cells (Table S6A). Additionally, we found that
transcripts from 2,145 and 3,259 genes were methylated only in
the mouse brain and HEK293T data sets, respectively, and that
transcripts from 2,509 genes were methylated in both data sets
(Table S6B). Interestingly, among the transcripts methylated in
regions of both orthologs (Figure 5E). Collectively, these data
indicate that m6A peaks are enriched near the stop codon in
Cell 149, 1635–1646, June 22, 2012 ª2012 Elsevier Inc. 1643
human transcripts and that many sites of methylation are
conserved in mouse and human transcripts.
Unlike DNA, which undergoes cytosine methylation and hydrox-
ymethylation, dynamic internal modifications of mRNA other
than RNA editing have not been established. Recent evidence
that the obesity risk gene, FTO, encodes a physiologic m6A
demethylase suggests that m6Ahas central rolesin cellular func-
tion. Here, we use MeRIP-Seq to provide the first transcriptome-
wide characterization of m6A. We show that m6A is a reversible
and widespread modification that is primarily located in evolu-
tionarily conserved regions and is particularly enriched near
the stop codon. We also find that many features of m6A localiza-
tion are conserved between the human and mouse transcrip-
tomes, and we uncover a previously unidentified link between
m6A and miRNA signaling. Collectively, these studies reveal
that m6A is a widespread and dynamically regulated base modi-
fication in mRNA, and they identify mRNAs that are most likely to
be influenced by signaling pathways that influence m6A levels.
One of the most striking features of m6A localization is its
prevalence within 30UTRs. The 30UTR is an important region
for RNA regulation, as it can influence RNA stability, subcellular
localization, and translation regulation. Several of these events
are regulated by RNA-binding proteins (RBPs) that bind to
cis-acting structural motifs or consensus sequences within the
30UTR and act to coordinate RNA processing. Conceivably,
m6A may influence the affinity of specific RBPs for their target
mRNAs, analogous to the recruitment of methyl-CpG-binding
et al., 1992). Given the abundance of m6A throughout the
transcriptome and its widespread localization, such a role for
m6A would be likely to have important consequences for the
regulation of numerous mRNAs.
Our profiling of m6A in HEK293T cells revealed thousands
of transcripts that are also methylated in the mouse brain. In
many cases, the patterns of m6A localization within these
transcripts are nearly identical, suggesting that some RNAs
possess highly conserved methylation profiles. However, we
also uncovered many transcripts that exhibit distinct cell-type-
specific methylation patterns, demonstrating that m6A is also
capable of being differentially regulated within unique cellular
Our finding that a large proportion of 30UTRs that contain m6A
peaks also contain miRNA binding sites is highly suggestive of
an association between m6A and miRNA function. Additionally,
our analysis indicated an inverse localization of m6A peaks and
miRNA-binding sites within 30UTRs, with m6A sites typically
preceding, but not overlapping, the miRNA sites in the 30
UTRs. Although miRNAs can inhibit their target mRNAs by
promoting either transcript degradation or translational repres-
sion (Guo et al., 2010; Hendrickson et al., 2009), the factors
thatdetermine whichfatepredominates arenotwellunderstood.
Conceivably, the proximity of m6A to a miRNA-binding site could
influence the mechanism of miRNA-mediated transcript inhibi-
tion. Additionally, it is possible that miRNA binding influences
m6A levels within 30UTRs. Indeed, our finding that abundant
miRNAs are more significantly enriched in m6A peaks than
weakly expressed miRNAs raises the possibility that miRNAs
regulate methylation status.
A surprising result of these studies is the finding that m6A is
highly enriched near stop codons. This recurrent localization
within transcripts suggests that adenosine methylation in the
vicinity of the stop codon may be of functional importance. Inter-
estingly, the consensus for adenosine methylation is relatively
short, and sequences that match the consensus are found
throughout the transcriptome. However, despite the frequency
of m6A consensus sites, methylation occurs primarily near stop
codons. It will be important to determine how this specificity is
achieved and whether cellular mechanisms that involve recogni-
tion of the stop codon or the beginning of the 30UTR are involved
in providing specificity to adenosine methylation.
The finding that FTO demethylates m6A suggests that mis-
regulation of pathways controlled by adenosine methylation
ultimately affect physiologic processes in humans. Although
m6A is found in many classes of RNA, it is intriguing to speculate
that FTO mutations mediate their effects by affected m6A in
mRNA. Indeed, our finding that FTO can demethylate diverse
mRNAs is consistent with this model. Direct characterization of
lish the mechanisms by which this mutation leads to disease.
In summary, our study demonstrates that m6A is a widespread
modification found in a large fraction of cellular mRNA. The
pervasive nature of this epitranscriptomic modification suggests
that adenosine methylation has important roles in RNA biology.
Much in the way that cytosine methylation and hydroxymethyla-
tion in DNA are important epigenetic regulators of the genome,
our data demonstrate that adenosine methylation in RNA is
a reversible modification throughout the transcriptome that is
Totalmousebrain RNA was subjected to RiboMinus treatment to reduce rRNA
contentasper themanufacturer’sinstructions(Invitrogen). RNAwasthenfrag-
mented to 100 nt sized fragments using Illumina Fragmentation Buffer accord-
ing to the manufacturer’s instructions and was subjected to two rounds of m6A
immunoprecipitation. Sequencing libraries were prepared using the Illumina
protocol for mRNA samples, and sequencing was performed on an Illumina
GAII3 or an Illumina HiSeq2000 as indicated. Genomic alignment (mm9 or
hg19 from UCSC genome browser) was done using the Burrows-Wheeler
Aligner (BWA) (Li and Durbin, 2010) at default settings, or using TopHat (see
Extended Experimental Procedures). We analyzed only those reads that (1)
uniquely mapped to the genome and (2) had a Phred quality score R20.
Additional methods are detailed in the Extended Experimental Procedures.
MeRIP-Seq data have been deposited in the GEO database under accession
figures, and six tables and can be found with this article online at doi:10.1016/
1644 Cell 149, 1635–1646, June 22, 2012 ª2012 Elsevier Inc.
We thank Dr. Richard Roberts and New England Biolabs for generously
providing the m6A (NEB) antibody. We also thank members of the Jaffrey
and Mason laboratories for helpful comments and suggestions. This work
was supported by NIH grants T32CA062948 (K.D.M.), 1F32MH095353-01
(K.D.M.), 1R01NS076465-01 (C.E.M.), and MH080420 (S.R.J.); NSF CAREER
grant 1054964 (O.E.); and the Starr Cancer Consortium (I4-A411; C.E.M.).
Received: December 12, 2011
Revised: March 15, 2012
Accepted: March 27, 2012
Published online: May 17, 2012
Beemon, K., and Keith, J. (1977). Localization of N6-methyladenosine in the
Rous sarcoma virus genome. J. Mol. Biol. 113, 165–179.
Bodi, Z., Button, J.D., Grierson, D., and Fray, R.G. (2010). Yeast targets for
mRNA methylation. Nucleic Acids Res. 38, 5327–5335.
Bringmann, P., and Lu ¨hrmann, R. (1987). Antibodies specific for N6-methyla-
denosine react with intact snRNPs U2 and U4/U6. FEBS Lett. 213, 309–315.
Brockman, J.M., Singh, P., Liu, D., Quinlan, S., Salisbury, J., and Graber, J.H.
(2005). PACdb: PolyA Cleavage Site and 30-UTR Database. Bioinformatics 21,
Cabili, M.N., Trapnell, C., Goff, L., Koziol, M., Tazon-Vega, B., Regev, A., and
Rinn, J.L. (2011). Integrative annotation of human large intergenic noncoding
RNAs reveals global properties and specific subclasses. Genes Dev. 25,
Canaani, D., Kahana, C., Lavi, S., and Groner, Y. (1979). Identification and
mapping of N6-methyladenosine containing sequences in simian virus 40
RNA. Nucleic Acids Res. 6, 2879–2899.
Carroll, S.M., Narayan, P., and Rottman, F.M. (1990). N6-methyladenosine
residues in an intron-specific region of prolactin pre-mRNA. Mol. Cell. Biol.
Church, C., Lee, S., Bagg, E.A., McTaggart, J.S., Deacon, R., Gerken, T., Lee,
A., Moir, L., Mecinovi? c, J., Quwailid, M.M., et al. (2009). A mouse model for the
metabolic effects of the human fat mass and obesity associated FTO gene.
PLoS Genet. 5, e1000599.
Church, C., Moir, L., McMurray, F., Girard, C., Banks, G.T., Teboul, L., Wells,
S., Bru ¨ning, J.C., Nolan, P.M., Ashcroft, F.M., and Cox, R.D. (2010).
Overexpression of Fto leads to increased food intake and results in obesity.
Nat. Genet. 42, 1086–1092.
ificity of mRNA N6-adenosine methyltransferase. J. Biol. Chem. 265, 20117–
Desrosiers, R., Friderici, K., and Rottman, F. (1974). Identification of methyl-
ated nucleosides in messenger RNA from Novikoff hepatoma cells. Proc.
Natl. Acad. Sci. USA 71, 3971–3975.
Dimock, K., and Stoltzfus, C.M. (1977). Sequence specificity of internal meth-
ylation in B77 avian sarcoma virus RNA subunits. Biochemistry 16, 471–478.
messenger RNA from cultured hamster cells. Nucleic Acids Res. 2, 1653–
Elemento, O., Slonim, N., and Tavazoie, S. (2007). A universal framework for
regulatory element discovery across all genomes and data types. Mol. Cell
Falnes, P.O., Klungland, A., and Alseth, I. (2007). Repair of methyl lesions in
DNA and RNA by oxidative demethylation. Neuroscience 145, 1222–1232.
Fawcett, K.A., and Barroso, I. (2010). The genetics of obesity: FTO leads the
way. Trends Genet. 26, 266–274.
Fischer, J., Koch, L., Emmerling, C., Vierkotten, J., Peters, T., Bru ¨ning, J.C.,
and Ru ¨ther, U. (2009). Inactivation of the Fto gene protects from obesity.
Nature 458, 894–898.
Gerken, T., Girard, C.A., Tung, Y.C., Webby, C.J., Saudek, V., Hewitson, K.S.,
Yeo, G.S., McDonough, M.A., Cunliffe, S., McNeill, L.A., et al. (2007). The
obesity-associated FTO gene encodes a 2-oxoglutarate-dependent nucleic
acid demethylase. Science 318, 1469–1472.
Guo, H., Ingolia, N.T., Weissman, J.S., and Bartel, D.P. (2010). Mammalian mi-
croRNAs predominantly act to decrease target mRNA levels. Nature 466,
Hendrickson, D.G., Hogan, D.J., McCullough, H.L., Myers, J.W., Herschlag,
D., Ferrell, J.E., and Brown, P.O. (2009). Concordant regulation of translation
and mRNA abundance for hundreds of targets of a human microRNA.
PLoS Biol. 7, e1000238.
Horowitz, S., Horowitz, A., Nilsen, T.W., Munns, T.W., and Rottman, F.M.
(1984). Mapping of N6-methyladenosine residues in bovine prolactin mRNA.
Proc. Natl. Acad. Sci. USA 81, 5667–5671.
Iwanami, Y., and Brown, G.M. (1968). Methylated bases of ribosomal ribonu-
cleic acid from HeLa cells. Arch. Biochem. Biophys. 126, 8–15.
tive demethylation of 3-methylthymine and 3-methyluracil in single-stranded
DNA and RNA by mouse and human FTO. FEBS Lett. 582, 3313–3319.
Jia, G., Fu, Y., Zhao, X., Dai, Q., Zheng, G., Yang, Y., Yi, C., Lindahl, T., Pan, T.,
Yang, Y.G., et al. (2011). N6-methyladenosine in nuclear RNA is a major
substrate of the obesity-associated FTO. Nat. Chem. Biol. 7, 885–887.
Klagsbrun, M. (1973). An evolutionary study of the methylation of transfer and
ribosomal ribonucleic acid in prokaryote and eukaryote organisms. J. Biol.
Chem. 248, 2612–2620.
Kong, H., Lin, L.F., Porter, N., Stickel, S., Byrd, D., Posfai, J., and Roberts, R.J.
(2000). Functional analysisofputativerestriction-modification systemgenes in
the Helicobacter pylori J99 genome. Nucleic Acids Res. 28, 3216–3223.
Lewis, J.D., Meehan, R.R., Henzel, W.J., Maurer-Fogy, I., Jeppesen, P., Klein,
F., and Bird, A. (1992). Purification, sequence, and cellular localization
of a novel chromosomal protein that binds to methylated DNA. Cell 69,
Lewis, B.P., Burge, C.B., and Bartel, D.P. (2005). Conserved seed pairing,
often flanked by adenosines, indicates that thousands of human genes are mi-
croRNA targets. Cell 120, 15–20.
Li, H., and Durbin, R. (2010). Fast and accurate long-read alignment with
Burrows-Wheeler transform. Bioinformatics 26, 589–595.
Licatalosi, D.D., Mele, A., Fak, J.J., Ule, J., Kayikci, M., Chi, S.W., Clark, T.A.,
Schweitzer, A.C., Blume, J.E., Wang, X., et al. (2008). HITS-CLIP yields
genome-wide insights into brain alternative RNA processing. Nature 456,
Munns, T.W., Liszewski, M.K., and Sims, H.F. (1977). Characterization of
antibodies specific for N6-methyladenosine and for 7-methylguanosine.
Biochemistry 16, 2163–2168.
Narayan, P., and Rottman, F.M. (1988). An in vitro system for accurate
methylation of internal adenosine residues in messenger RNA. Science 242,
Pang, K.C., Stephen, S., Engstro ¨m, P.G., Tajul-Arifin, K., Chen, W., Wahles-
tedt, C., Lenhard, B., Hayashizaki, Y., and Mattick, J.S. (2005). RNAdb—
a comprehensive mammalian noncoding RNA database. Nucleic Acids Res.
33 (Database issue), D125–D130.
Perry, R.P., Kelley, D.E., Friderici, K., and Rottman, F. (1975). The methylated
constituents of L cell messenger RNA: evidence for an unusual cluster at the
50terminus. Cell 4, 387–394.
Pollard, K.S., Hubisz, M.J., Rosenbloom, K.R., and Siepel, A.(2010).Detection
of nonneutral substitution rates on mammalian phylogenies. Genome Res. 20,
Proudfoot, N. (1991). Poly(A) signals. Cell 64, 671–674.
Cell 149, 1635–1646, June 22, 2012 ª2012 Elsevier Inc. 1645
Saneyoshi, M., Harada, F., and Nishimura, S. (1969). Isolation and character-
ization of N6-methyladenosine from Escherichia coli valine transfer RNA.
Biochim. Biophys. Acta 190, 264–273.
Stoltzfus, C.M., and Dane, R.W. (1982). Accumulation of spliced avian retro-
virus mRNA is inhibited in S-adenosylmethionine-depleted chicken embryo
fibroblasts. J. Virol. 42, 918–931.
Tian, B., Hu, J., Zhang, H., and Lutz, C.S. (2005). A large-scale analysis of
mRNA polyadenylation of human and mouse genes. Nucleic Acids Res. 33,
Wei, C.M., Gershowitz, A., and Moss, B. (1975). Methylated nucleotides block
50terminus of HeLa cell messenger RNA. Cell 4, 379–386.
Wei, C.M., Gershowitz, A., and Moss, B. (1976). 50-Terminal and internal meth-
ylated nucleotide sequences in HeLa cell mRNA. Biochemistry 15, 397–401.
1646 Cell 149, 1635–1646, June 22, 2012 ª2012 Elsevier Inc.