m 6 A mRNA modifications are deposited in nascent pre-mRNA and are not required for splicing but do specify cytoplasmic turnover

1Studying nascent RNA synthesis in cultured cells using very brief pulse labeling with radioactive nucleosides allowed a number of advances in understanding pre-mRNA synthesis and processing in the era before rapid nucleic acid sequencing. Examples include polyA addition on pre-mRNA before completion of mRNA processing and cytoplasmic entry and locating the first known boundaries of eukaryotic polymerase II transcription units through studying labeled nascent adenovirus transcripts.

2These early experiments were joined by a cell fractionation technique originated by that uses a 1 M urea solution to liberate a “chromatin” fraction from nuclei. This fraction provides a stringent purification of growing nascent pre-mRNA chains, isolated as a chromatin-associated RNA fraction (referred to as CA-RNA), along with RNA polymerase II plus all nuclear DNA and associated histones. Using specific labeled DNA probes, demonstrated removal in liver cell nuclei of some, but not all, introns from two specific nascent pre-mRNAs: a transcription factor pre-mRNA involved in circadian rhythm and the HMG coA reductase pre-mRNA. Recently, adapted this procedure to study the extent and order of intron removal in cultured human carcinoma cell nuclei, again showing that many, but not all, introns are removed in CA-RNA.

3Current sequencing techniques allow a complete analysis of the three cellular RNA fractions: CA-RNA, nucleoplasm, and cytoplasm. and recently reported sequencing analyses of these three RNA fractions at various times after gene induction of toxin-responsive genes in macrophages. Major conclusions were that (1) many, but not all, introns are removed from the pre-mRNA of the induced genes in the CA-RNA during pre-mRNA processing and prior to transfer to the nucleoplasm, by which time essentially all introns had been removed, as observed for the cytoplasmic RNA. (2) Individual intron removal time varied greatly from a few to many minutes in different individual induced and constitutive mRNAs. The order of removal generally favored 5′-to-3′ removal, but this was definitely not universally true. (3) All analyzed chains had the 3′ polyA added before completing intron removal and departure from the chromatin fraction. (4) Within minutes of the appearance of these specific induced mRNAs in the nucleoplasm, they appeared in the cytoplasm. In an independent study,, ) also used this nascent chain technique coupled with sequencing in mouse cells and Drosophila tissue to demonstrate partial intron removal in CA-RNA.

4Examination of RNA separated into these three fractions is vastly more informative concerning events in the synthesis and processing of pre-mRNA than the now almost universally performed extraction of total RNAs or examination of crude whole nuclear preparations.

5Methylation on the N6 position of adenosine in eukaryotic polyA+ mRNA from polysomes was discovered in cultured mammalian cells in 1974 followed by many additional studies, including demonstration of N6-methyladenosine (m6A) in high molecular nuclear DNA-like RNA (hnRNA), which contains pre-mRNA.

6Since the development of a highly specific antibody to m6A, now commercially available, interest has returned recently to the study of this mRNA modification with considerable progress. Use of m6A antibody to precipitate oligonucleotides derived from total mammalian RNAs followed by sequencing revealed a general distribution of m6A in sequences that map to mRNAs and was reported as enriched around stop codons. Site-specific localization of m6A by m6A cross-linking immunoprecipitation (m6A-CLIP, modifying the CLIP technique) demonstrated a precise enrichment of m6A in the start of 3′ exons but not around stop codons. Rather, m6A modification in last exons occurs more in terminal exons harboring 3′ untranslated regions (UTRs) than in coding sequences (CDSs) in both cultured human cells and mouse brains and livers, allowing possible 3′ UTR regulation, including alternative polyadenylation. Some recent experiments have begun to shed light on specific individual m6A additions to mRNAs with 5′ UTR that promote non-cap-dependent translation in the 5′ UTR. This work deals with “stress” (heat shock) mRNAs and is thought to depend on the affinity of EIF3 for m6A. A separate clear finding shows that m6Am on the first A residue in mRNAs inside the m7G of the cap structure at the 5′ end of mRNAs protects specific RNA from turnover. Other studies have attempted to link m6A to generalized mRNA turnover, translation efficiency, splicing, and alternative polyadenylation.

7In this study, we first determined whether m6A is added to CA-RNA in the nucleus during pre-mRNA synthesis and processing and found that this methylation occurs in exonic but not intronic sequences in CA-RNA. We then compared the individual m6A present in HeLa cell pre-mRNA in CA-RNAs with the m6A content and location in the nucleoplasmic and cytoplasmic mRNAs and found it to be the same. One role of m6A in mRNA, first suggested in 1978 and recently suggested by, is to contribute to faster mRNA turnover. We used m6A-CLIP sequencing to quantitate the contribution to mRNA turnover in HeLa cells, normal mouse embryonic stem cells (ESCs), and Mettl3 knockout mouse ESCs. These results demonstrate that while m6A deposition occurs on nascent RNA in exons, it is not required for splicing but decreases cytoplasmic mRNA for those transcripts.

8To compare m6A deposition in different cellular compartments, three RNA fractions—CA-RNA, nucleoplasm, and cytoplasm—were prepared from HeLa cells for sequence analysis, m6A location, and quantification (Fig. 1; Supplemental Fig. 1).

9Sequence analysis of the three HeLa cell fractions (Fig. 1A) showed that ∼65% of the exons of nascent transcripts contained various numbers of introns in the CA-RNA fraction along with 35% of the exons of CA-RNA transcripts that were completely spliced. This agrees with earlier studies in other cells. We proceeded to precisely map m6A using the m6A-CLIP method. Isolated RNA was fragmented to ∼80 nucleotides (nt) followed by precipitation of m6A-containing oligonucleotides by m6A-specific antibody, cross-linking by UV irradiation of the bound RNA to antibody, collection of the bound complex, and release of methylated RNA by protein digestion followed by cDNA library construction and RNA sequencing (RNA-seq). Reverse transcription of RNA containing an m6A cross-linked peptide/amino acid or even m6A modification itself occasionally results in detectable diagnostic sequence errors at the m6A sites, including cross-linking-induced mutation sites (CIMS; such as single-base substitutions, deletions, and insertions), cross-linking-induced truncation sites (CITSs), and m6A-induced truncation sites (MITSs). These changes allow precise identification of the sites of m6A modification. The m6A antibody-enriched RNA regions are quantified as m6A peak regions or m6A peaks, within which the consensus m6A site at the mutational or truncation site precisely identifies m6A location. We included in the Supplemental Material a protocol for m6A-CLIP with a detailed day-to-day arrangement of experiments.

10Given that HeLa cell CA-RNA retains many introns (Fig. 1A), a major initial result was that ∼93% of the m6As in the partially spliced CA-RNAs were in exons (Fig. 1B; Supplemental Fig. 2) despite the fact that there was about three times as much intronic as exonic sequence within the partially spliced CA-RNA (Fig. 1C). In both the nucleoplasm and cytoplasm, ∼98% of the m6As were identified in exons. Therefore, an A residue in an intron has a small chance of being methylated compared with an A residue in an exon.

11In our previous work, we found a large fraction (∼70%) of m6As in total steady-state mRNA in cultured cells and tissues to be in last exons, about two-thirds of which are in the 3′ UTR sequence before the 3′ polyA terminus of an mRNA.

12Our finding that m6A is added mostly to exons, not introns, in CA-RNA prompted us to compare the previously noted peak distribution of m6A within the steady-state 3′-terminal exons in total cell RNA with that in each of the cell fractions. Both the specific sequence location (Fig. 1D; Supplemental Fig. 3) and the modification level (Fig. 2A,B, each dot is an individual m6A peak) of each m6A peak in CA-RNA were quantitatively very similar to that in the nucleoplasmic and cytoplasmic fractions. This result suggests that the methyltransferase complex plus associated proteins must deposit m6A soon after chain synthesis regardless of whether splicing has occurred and that there is no frequent methylation/demethylation occurring past the pre-mRNA stage (∼90% of m6A peaks are unchanged when comparing CA-RNA and nucleoplasmic RNA [Fig. 2A], and >99% of 37,557 m6A peaks are unchanged when comparing nucleoplasmic and cytoplasmic mRNA peaks [Fig. 2B]).

13We included an analysis of the ∼5% of primary transcripts that, when processed, do not have polyA added to the last coding exon but instead have it added to a 3′-terminal noncoding exon (Fig. 1D; Supplemental Fig. 3, red lines). The nascent chromosomal RNA of these particular transcripts also received a high level of methylation in the terminal exon, as did the great majority of conventional mRNAs (Fig. 1D; Supplemental Fig. 3, black line). This reconfirms that the actual enrichment of m6A in last exon is established while pre-mRNA is still attached to chromatin.

14In the experiments of, it was established that cleavage and polyA addition were completed in most pre-mRNAs that are still associated with chromatin before splicing was completed. The present results show that m6A also is added to pre-mRNA before mRNA leaves the chromatin in what appears to be virtually all of the same modifications subsequently found in nucleoplasmic and cytoplasmic RNA.

15To gain insight into m6A addition as a function of position of the polymerase along the gene, further analysis of the distribution in the CA-RNA was carried out (Supplemental Fig. 4). Internal exons (excluding first and last exons) on nascent chains were divided into four groups according to the distance of each exon to a polyA site: quartile 4 (Q4; the longest) and Q1 (the shortest). Splicing completion and the frequency of m6A in internal exons were assessed in each quartile (Supplemental Fig. 4). For example, Q1 contained nascent chains that were up to ∼6.4 kb, and Q4 contained the largest nascent chains, up to ∼1 Mb (Supplemental Fig. 4A). To calibrate the splicing completeness of internal exons on CA-RNAs, we calculated a splicing completion index for each internal exon in CA-RNA and nucleoplasmic mRNA, where the splicing completion index was defined as the percentage of spliced RNA molecules over the total RNA molecules, including both spliced and unspliced. The index difference for each exon between CA-RNA and nucleoplasmic mRNA measures the splicing completeness: The higher the difference, the less complete the splicing is. As described previously, completion of splicing showed that the most 5′ –proximal internal exons were spliced more completely than the exons close to the polyA site (Supplemental Fig. 4A). Strikingly, exons in all four groups of nascent RNA had a similar frequency of exonic m6A residues (∼4% of exons in each group had at least one m6A) (Supplemental Fig. 4B). These results suggest that m6A addition can occur on exons in nascent chains regardless of whether splicing of the exons has occurred.

16In the course of sample preparations to study m6A locations, RNA was reduced to fragments nominally 50–80 nt long, cross-linked to m6A antibody, reacted with m6A-specific antiserum, precipitated, and sequenced. If some of the m6A-containing fragments contained exon–intron junction sequences within the CA-RNA fraction, they should be greatly enriched by the m6A antibody precipitation. When we assessed this, we found that the m6A containing exon–intron junction sequence fragments were very uncommon but were significantly enriched after m6A antibody precipitation (Fig. 3A, left) compared with exon–intron junction sequence fragments lacking m6A (Fig. 3A, right). Two examples of junction sequences with an m6A still retaining exon–intron junctions are shown in Figure 3, B and C. The m6A is near a 5′ splice site in one and a 3′ splice site in the other. (It is important to recognize that this result does not comment on the density of m6A at splice sites; rather, it shows only that m6A addition has occurred before splicing is complete. The density of m6A around splice sites is discussed later.) Four more such examples are shown in Supplemental Figure 5. More than 200 cases were examined in which sequenced internal exons have m6A-containing exon–intron junction fragments (Supplemental Table 1); the m6A site in every examined exon–intron junction fragment from CA-RNA was within the exon, proving that methylation can (and apparently frequently does) occur before splicing is completed and reinforcing the conclusion that m6A addition is to exons but not introns. These observations strongly suggest that m6A is added to exons coincident with or soon after exon definition occurs in nascent pre-mRNA.

17With respect to a relation between m6A addition and splicing, two extreme possibilities exist: (1) m6A addition to an exon is independent of and not involved in intron removal or (2) intron removal depends on or is blocked by m6A sites in exons bordering introns.

18As has been known for more than two decades, the favored methylation sequence for m6A modification in mRNA is RRACU or at least RAC. The short consensus at 5′ and 3′ splice sites (5′AG/GT --------NAG/GNN3′; no strict conservation) in higher eukaryotic RNA splicing seems to rule out any direct marking of splice sites by m6A in either directly locating splice sites or, alternatively, directly blocking splice sites.

19Whether m6A might be located in an exonic splicing enhancer or exonic splicing silencer sequence cannot be determined directly from total sequence compilations (e.g., Supplemental Table 2), but the position of m6A residues with respect to intron–exon boundaries can be assayed. We therefore determined the proximity of m6As to splice sites. This positional assessment was carried out in all three RNA fractions: CA-RNA, nucleoplasmic RNA, and cytoplasmic RNA. Starting at either a 5′ or 3′ splice site and moving into the exon sequence, the m6A occurrence rises so that by ∼100 nt, the density equals the average density in all internal exonic regions (Fig. 4A). The occurrence in the first 50 nt away from the splice junction (3′ or 5′) is only half the average occurrence of that in internal exonic regions (Fig. 4A,B). Furthermore, only 7% of m6As are within 50 nt of splice sites in these internal exons (Fig. 4B, top left), although this proximal splice site region contains 13% of total RRACUs in the internal exons (i.e., ∼50% less of the RRm6ACU motif compared with the RRACU motif for the 50-nt region near the splice site region; P < 10−24, Fisher's exact test) (Fig. 4B, top right).

20We also noted that certain internal exons show relative enrichment of m6As: Long exons (≥200 nt) contain 80% of all internal exon m6As while containing only 28% of total (modified + unmodified) internal exon RRACU motifs (i.e., >10-fold enrichment of RRm6ACU compared with the RRACU motif in exons >200 nt; P < 10−100, Fisher's exact test) (Fig. 4B, bottom two panels).

21In the total CA-RNA, there was an average of ∼10%–15% more m6A in the 100 nt near both the 3′ or 5′ splice junctions than the same regions in either the nucleoplasmic or cytoplasmic RNA (Fig. 4A). Furthermore, we found that while ∼30,000 m6A peaks are present at the same frequency in chromatin and nucleoplasmic fractions, ∼4000 individual m6A peaks had a higher frequency in CA-RNA compared with nucleoplasmic RNA (Fig. 2A). The majority (91%) of the m6As that are higher in CA-RNA is at least 50 nt away from splice sites (Fig. 4D). These modifications are preferentially located in internal constitutive exons (Supplemental Fig. 6). Those m6A peaks that are higher in CA-RNA show a 50% higher frequency in both 5′ and 3′ junctions in exons (Fig. 4C). However, this m6A enrichment in exons near splice junctions in the CA-RNA of individual pre-mRNAs does not persist in nucleoplasmic or cytoplasmic RNA of these specific transcripts (Fig. 2A). It seems possible that these m6A residues are removed by demethylases prior to nuclear exit.

22We also examined specifically the residues at the 5′ and 3′ borders of pre-mRNAs that harbor alternatively spliced exons. As was true for 5′ and 3′ splice sites, in general, there was no accumulation of m6A residues located at splice sites of these alternatively spliced exons (Supplemental Fig. 7A). We also examined sequences bordering alternatively spliced exons in mouse ESCs (discussed fully later; see Supplemental Fig. 7B) and likewise found no heightened collection of m6A residues at exon–intron junctions.

23This is in sharp contrast to recent reports claiming multiple distinct deposits of m6A at both 5′ and 3′ splice sites of thousands of mRNAs in HeLa and 3T3 cells. Hence, we reanalyzed raw sequencing data from different laboratories (Supplemental Fig. 8) and were unable to reproduce the findings of m6A heightened at splice sites but were able to reproduce results from two independent groups (the Regev group and the Jaffrey group) that were consistent with ours. In summary, the data indicate that the bulk of m6A modifications is located internally in exons and that there is no enrichment of m6A at splice junctions, while there is no absolute prohibition of m6A sites within 50 nt of splice sites (see Fig. 4; Supplemental Figs. 7, 8).

24We next quantitatively examined several properties of mRNA from growing normal ESCs and the same cell type with a knockout of the Mettl3 gene. Mettl3 was the first cloned mRNA m6A methyltransferase and is the major methyltransferase in a larger methylation complex. A recent study established that the Mettl3/Mettl14/WTAP protein complex is necessary for the m6A methylation of mRNAs in A549 cells. Mettl3 knockout cells no longer differentiate upon stimulation but continue to grow in appropriate medium even though the mRNA m6A content is ∼10% of normal (see Fig. 6A, below). We first examined the effects on RNA profiles and splicing that might be caused by Mettl3 deletion. The profiles of the mRNAs in the two cell samples were largely overlapping, and, with a statistical cutoff commonly used in splicing research for reliable splicing detection (false discovery rate [FDR] <5% and ΔPSI [percent spliced in] ≥0.1), all m6A-containing constitutive exons (>8000 constitutive exons) were spliced quantitatively the same in wild-type and knockout cells (Fig. 4E).

25We also examined ∼2000 alternately spliced cassette exons containing m6As in the presence or absence of Mettl3. In Mettl3 knockout cells, only 4% of these alternatively spliced cassette exons showed more frequent inclusion, and 1% showed less frequent inclusion of the alternative exon (where splicing changes were defined as FDR <5% and ΔPSI ≥0.1) (Fig. 4F). We also examined the splicing for all constitutive exons and alternative exons regardless of whether they contained m6A (i.e., even considering the indirect effects of m6A on splicing). Again, >67,000 constitutive exons were spliced identically, and only a very minor proportion (∼3%) of ∼12,000 alternative cassette exons changed splicing (see details in the legend for Fig. 4F). Thus, after the removal of >90% of m6A by the knockout of Mettl3, there is no detectable effect on constitutive exons and a minor effect on mRNA splicing of alternative exons containing m6A. Furthermore, we downloaded the raw RNA-seq data from previous publications that reported certain splicing changes after Mettl3 knockdown, performed the same splicing analysis commonly used in splicing research for reliable splicing detection, and observed essentially the same result as these published data with our data: Loss of m6A has no detectable effect on constitutive splicing and a minor effect on the splicing of alternative exons with m6A modification (Supplemental Fig. 9).

26Mammalian cell mRNAs were shown decades ago to have half-lives (T1/2s) varying from ∼1 h or less to ≥12 h. Furthermore, the most rapidly labeled mRNAs are enriched in short-lived mRNAs compared with mRNAs labeled for 24 h. Labeled nascent RNA reflects “rate of synthesis”; stable RNAs in cytoplasm will certainly be labeled longer than unstable RNAs that are labeled briefly in steady state. The briefly labeled mRNAs were also found to be enriched 2.5-fold in m6As compared with stable mRNAs. Recent reports using sequence analysis have now also suggested a link between m6A content and mRNA T1/2. Extensive T1/2 measurements for individual sequenced mammalian mRNAs that are available after actinomycin D treatment of cultured cells suggest a wide spectrum of T1/2s, from as little as ≤1 h to ≥10 h.

27We examined available T1/2 data for ∼8000 individual HeLa cell mRNAs. The mRNAs were divided into four quartiles (Q1–Q4; fastest to slowest turnover). m6A content of individual mRNAs in each quartile was then determined and plotted. Figure 5A shows m6A content in each quartile plotted on the Y-axis by position in the mRNA chains, anchored at start and stop codons. The density of m6A, normalized for length, among all exons was greatest in the coding region and the 3′ UTR of those mRNAs (anchoring at last exon start yields the same result) (Fig. 5A; Supplemental Fig. 10A). These most unstable human cell RNAs (Q1) have approximately two to four times more m6A than the most stable (Q4) HeLa cell mRNAs in exons of both the coding region and the 3′ UTR. Similar associations of more m6As in mRNAs with shorter T1/2s were obtained using the earlier T1/2 data of our group from human hepatoma cell (Supplemental Fig. 10B; ). We checked whether there is a correlation between the occurrence of m6A in the CDS versus the 3′ UTR of the same RNA and found none (R2 = 0.01 on a transcript basis); thus, m6A in both the CDS and the 3′ UTR specifies mRNA turnover independently.

28We next analyzed the location of m6A in transcripts showing changes in T1/2s in wild-type versus Mettl3 knockout ESCs. We first determined the T1/2s of individual polyA+ mRNAs in wild-type and Mettl3 knockout ESCs by sequencing polyA+ mRNA at various times after actinomycin D treatment. The T1/2s of individual polyA+ mRNAs were reproducible in biological triplicates, and the integrated data of individual replicates were used in the following analyses. Knowing the T1/2s of many mRNAs, we then determined the m6A content of all of these mRNAs. As in HeLa cells, the mRNAs with the shortest T1/2s had the highest m6A density (Fig. 5B; Supplemental Fig. 10C). Also, as in HeLa cells, the highest frequency of m6A is in the last exons, the CDSs, and the 3′ UTRs (Fig. 5B; Supplemental Fig. 10C).

29With the T1/2 data and the m6A information of individual mRNAs, it was possible to directly relate the two (Fig. 5C). A plot of the number of m6As as a function of mRNA T1/2 in individual mRNAs emphasized strongly the high correlation between m6A content and shorter T1/2s (Fig. 5C, left panel). Many mRNAs with a T1/2 of 5 h or less had 10 m6As and some had even more. mRNAs with T1/2s of 10–20 h or >20 h had a progressively smaller number of m6A peaks.

30The distribution of individual mRNAs with no, one, or multiple m6A peaks is shown in Figure 5, D and E. The quartile of mRNAs with the shortest T1/2 had only ∼10%–12% mRNAs with no m6A, whereas the quartile with the longest T1/2 had almost 40% of individual mRNAs with no m6A (Fig. 5D). Furthermore, the mRNAs with the shortest T1/2 had ∼75% of individual mRNAs with two or more m6As compared with 25% in mRNAs with the longest T1/2. The mRNA T1/2 distribution showed that mRNAs with multiple m6As have a median T1/2 less than half that of mRNAs with no m6A (Fig. 5E).

31However, it is important to point out that many mRNAs lacking m6A still have a T1/2 of ≤5 h, and some of these mRNAs have a T1/2 of only 1 or 2 h (Fig. 5E). Clearly, there are triggers for turnover other than m6A. Compared with mRNAs with no m6As, mRNAs with m6As in the 3′ UTR have a faster median T1/2 (Fig. 5F), and the T1/2 is even faster for mRNAs with m6A in the CDS. The group of mRNAs with m6As in both the CDS and the 3′ UTR have the fastest median T1/2, almost double that of mRNAs with no m6A (Fig. 5F). Finally, the distance between individual m6A peaks in the 25% of mRNAs with the shortest T1/2 was less than the 50% of mRNAs with the longest T1/2 (Fig. 5G). This might mean it will eventually be found that clustered m6As function in initiating rapid mRNA turnover.

32The Mettl3 knockout in ESCs resulted in a great reduction in m6A in the polyA+ mRNA (90%) but had no detectable effect on methylation of other cellular RNAs (i.e., polyA− RNA, most of which is rRNA) (Fig. 6A; Supplemental Fig. 11). The effects of Mettl3 knockout showed a fourfold to eightfold increase in T1/2 in many of the individual mRNAs regardless of whether the m6A peak loss was for the total mRNA or in the CDS or 3′ UTR (Fig. 6B). The quantitative effect was slightly greater when the m6A was in the CDS compared with the 3′ UTR, and simultaneous m6A losses in both the CDS and the 3′ UTR gave the maximum effect in increasing T1/2 (Fig. 6C). This longer T1/2 in the knockout cells was observed for mRNAs over the whole range of RNA T1/2s (Fig. 6B,E) and correlated with the m6A content; that is, multiple m6A loss increased T1/2 the greatest (Fig. 6B,E). The increased T1/2 related to m6A loss resulted in increased steady-state levels of the target mRNAs (Fig. 6F). We also noted that mRNAs with the greatest increase in T1/2s had more closely spaced m6A peaks than the m6A peaks in mRNAs with a smaller effect on the T1/2 of the knockout (Fig. 6G).

33Although these results with both HeLa cells and mouse ESCs show a definite qualitative and quantitative contribution of m6A to mRNA turnover, every single m6A is not an obligatory signal for rapid turnover.

34Most mRNAs that encode key proteins in controlling the ESC state have multiple m6As and short T1/2s (Fig. 7A). As has been reported with fewer examples, gene ontology (GO) analysis revealed that T1/2 mRNAs with faster turnover and multiple m6As are associated with regulatory functions, such as transcriptional regulation (Fig. 7A,B), while more stable mRNAs with no m6As are associated with housekeeping functions, as illustrated by structural parts of the ribosome (Fig. 7B; Supplemental Tables 3, 4). Interestingly, mRNAs with a short T1/2 and no m6A are associated with nucleosome-related functions, such as nucleosome-related proteins (Fig. 7B; Supplemental Table 5).

35The earlier conclusion that mRNA released into the nucleoplasmic fraction from CA-RNA is completely processed with respect to splicing and polyA addition is corroborated and extended in the present work. Here we found that virtually all m6A methylation occurs at the pre-mRNA level during transcription and mRNA processing and is completed by the time an mRNA is released from chromatin into the nucleoplasmic RNA. Perhaps the most far-reaching new finding is that the vast majority of m6A addition occurs in exon sequences within CA-RNA—demonstrably, in many cases, before splicing has occurred—and not in introns even though the introns have many more total m6A target sequences (RAC or RRACU). Thanks to the precise identification of m6A sites afforded by m6A-CLIP, we conclude that the same m6A modifications (in both position [Fig. 1D] and robustness [Fig. 2A,B] for each m6A) in pre-mRNA are in the nucleoplasm and, finally, the cytoplasm. Thus, m6A modifications in pre-mRNA (CA-RNA) are not necessary for splicing but determine the subsequent role of m6A in cytoplasmic mRNA T1/2.

36Several important conclusions follow from the results above. First, the proteins in the 800-kDa complex originally identified as necessary for in vitro methylation of mRNA plus the m6A enzyme complex itself somehow distinguish between exon and intron as the nascent chain is being synthesized and before a great many introns are removed.

37A similar conclusion was described in 3H uridine- and 3H methyl methionine-labeling experiments of late adenovirus mRNA formation. There, ∼14 different mRNAs are fashioned from a single 28-kb primary transcript (for review, see ) that is always completed even though processing (polyA addition) begins before chain completion. A kinetic analysis showed a greater conservation of 3H methyl label in pre-mRNA in the formation of adenoviral mRNA compared with the lesser conservation of 3H uridine. This early result implied that m6A is added only to that portion of the 28-kb primary transcript destined to be retained in one of the ∼14 different 2- to 4-kb mRNAs.

38Thus, a crucial open question begs an answer: How does the m6A methylation complex know which newly formed pre-mRNA sequences will survive processing into mRNA? Put another way, how does m6A modification in pre-mRNAs occur with respect to “exon definition”? Recent experiments on the regulation of splicing have used a fractionation procedure that produced a high-molecular-weight (HMW) nuclear fraction that was released from chromatin and preserved some protein–protein interactions. We tested whether antibody precipitation of the proteins in the HMW fraction would connect m6A methyltransferases and U1 snRNP and U2AF proteins, since exon definition is often hypothesized to depend on the initial U1 RNA binding at the 5′ splice site and U2AF at the 3′ splice site. In our hands, methyltransferase antibodies (Mettl3 or Mettl14) did not precipitate U1 snRNP or U2AF proteins, and U1 snRNP and U2AF protein antibodies did not precipitate Mettl3 or Mettl14 (Supplemental Fig. 12). A previous study reported a similar finding that U1 or U2 snRNP precipitation by antisense oligos could not pull down either Mettl3 or Mettl14.

39Another important implication arises from the fact that m6A modification is the same in the newly formed pre-mRNA as in the nucleoplasmic and steady-state cytoplasmic mRNAs (Fig. 7C, the m6A information flow from chromatin to nucleoplasm and cytoplasm). This result argues strongly against a primary role for “dynamic” regulation of methylation and demethylation that has been proposed.

40With respect to splicing, the present work showed that ∼90% of the m6As in the internal coding exons are not within 50 nt of splice sites, and Mettl3 knockout cells have very few changes in either constitutive or alternative splicing. Furthermore, the Mettl3 knockout mouse embryonic cells make and splice mostly the same mRNAs as wild-type cells. Therefore, m6A is not obligatory for most splicing. It remains possible that a minority of m6As is involved in splicing choices.

41Is the completion of splicing the limiting factor for release of a completed mRNA molecule from the chromatin? Since many introns are removed from nascent chains before as well as after a polyA tail has been added and both polyadenylation and m6A methylation are apparently completed on CA-RNA, it would appear that completion of splicing could be the limiting step in “release.” Early work in mouse erythroleukemia cells showed that β-globin chains mutant in either splice sites or polyA sites never leave the chromatin. Thus, effective processing is required to release RNA from the site of transcription.

42At least one other potentially time-consuming step to assure mRNA quality control and T1/2 in the cytoplasm is binding of exon junction complexes (EJCs). An antibody study of the presence or absence of EJCs on CA-RNA in comparison with nucleoplasmic RNA would be most helpful in this regard. A related idea is that the results of m6A addition point to the possibility that recruitment of the 800-kDa complex and methyltransferases to nascent pre-mRNA is similar to the proposed recruitment of splicing machinery and exon definition within pre-mRNA. This possibility must be assessed biochemically.

43Our present experiments comparing normal mouse embryonic cells and those with no Mettl3 reveal a clear picture that at least many of the site-specific m6As in the CDS and the 3′ UTR function in governing mRNA T1/2. Identification of specific m6As that trigger specific mRNA turnover is the key to studying at least the biochemistry of proteins/protein complexes involved in the turnover of specific m6A-containing mRNAs. To take one evident case: mRNAs with short T1/2s can have easily ≥10 m6A peaks (clusters). Do these multiple m6As act in concert? Furthermore, since ∼20% of mRNAs with no m6As do still turn over more rapidly than others without m6A, what is the trigger for their turnover?

44To begin to analyze specific m6As and mRNA T1/2, we linked the CDS (GFP) in-frame to mouse ESC segments of two different mRNA regions rich in m6A. We used both wild-type and mutated sites that removed the normally methylated A. This deletion produced composite mRNAs with slower T1/2s in the mutants than constructs containing the wild-type sequences (Fig. 7D–K). It was true, however, that the T1/2 difference in the mutant construct did not affect T1/2 equally with the same wild-type sequences (Fig. 7D–K).

45In other recent experiments, m6A-directed RNA binding to specific protein domains of the YTHDF family, which do bind m6A-containing oligonucleotides, was at first thought to be involved in mRNA functions such as translation or splicing. The m6A methylated RNA binding to a motif of YTHDF2 was first used by. The YTHDF2 RNA-binding protein was then used to collect m6A-containing mRNAs for study. An siRNA knockdown of this protein resulted in ∼20%–30% slower turnover for the RNAs that bound most avidly to YTHDF2. A following study also concluded, based on knockdown experiments of the YTHDF1 protein, that “translation efficiency” was dependent on m6As in mRNA. On the basis of these results, summary reviews were published proposing that general roles for m6A had been established in mRNA turnover and mRNA translational regulation and that addition and removal of m6A residues were “dynamic”. A very recent work reiterated the idea that m6A in mRNA is subject to a “dynamic” cycle. Such a conclusion implies removal and addition in the cytoplasm of m6A at specific sites. However, as noted above, our results present a strong challenge to this idea of a dynamic cycle that involves a large fraction of the mRNAs in the cells that we examined. Without a doubt, there is a physiologic effect of demethylases, but the notion of easy frequent reversibility of m6A addition/subtraction, at least in steady-state tissue culture cells, is no longer logical. The m6A methylation occurs to nascent pre-mRNA, and the present data argue that it is largely stable in the cytoplasm until the mRNA turnover. If it were to occur, our data indicate that cycles of addition and turnover would have to be targeted to the same sequences within the mature mRNA, for which we found no evidence. Furthermore, it remains possible that methylation removal by demethylases might occur only within the nucleus on CA-RNA transitioning into the nucleoplasm (see Fig. 2A, where specific m6A residues in CA-RNA were removed during the release to the nucleoplasm before exit to the cytoplasm.).

46We recently found that m6A density is maximal early in the 3′ UTR, and at least some of these m6As appear to block use of more upstream polyA sites, thereby resulting in utilization of downstream polyA sites. In all likelihood, the large number of m6A sites in the CDSs and 3′ UTRs might affect a variety of functions, including mRNA stability.

47A recent breakthrough has clearly established a function of m6A addition to some specific sites near the 5′ end of the mRNA. Eukaryotic initiation factor EIF3 can bind m6A that has been added in the 5′ UTRs of mRNA and thereby promote non-cap-mediated translation initiation. Evidence of whether this 5′ UTR m6A addition on “stress” mRNAs occurs in the nucleus on pre-mRNA on the way to produce a new mRNA or whether the methylation occurs in the cytoplasm to pre-existing mRNA was not provided. The examination of m6As in nascent CA-RNA for these particular mRNAs under stress conditions is now strongly suggested.

48Finally, the existence of m6A in A residues in the 5′ cap of mRNA (m6Am) was first reported in the 1970s. When the first nucleotide in a pre-mRNA is an A residue, many such A residues acquire m6Am modification. A recent finding indicates that such mRNAs are generally long-lived and are demethylated by the FTO demethylase. Overexpression of FTO causes a faster turnover of these mRNAs, and an FTO knockout stabilizes these same mRNAs. As capping occurs shortly after pre-mRNA initiation by enzymes known to act in the nucleus, these results almost certainly mean that m6Am modification occurs at least initially in CA-RNA. These results also provide a connection between m6Am modification and subsequent effects on mRNA T1/2 but suggest that a different mechanism is at play when the 5′ end of the transcript is capped and methylated compared with internal m6A modification.

49In summary, use of the widely available highly specific m6A antisera plus the availability of low-cost sequencing and now the application of an updated cell fractionation protocol have quickly advanced our knowledge of a decades-old problem of m6A—the site-specific location and the function of m6A in mRNA. We now know that the specific sites found in mRNA are methylated on the nascent pre-mRNA and can link this methylation to at least several specific functions of some of the m6A residues in eukaryotic mRNA.

50HeLa cells were grown to 90% confluency in DMEM (Life Technologies) supplemented with 10% fetal bovine serum (Omega), 2 mM L-glutamine (Life Technologies), 1 U/mL penicillin/streptomycin (Life Technologies), 1× nonessential amino acids (Life Technologies), and 0.1 mM β-mercaptoethanol (Sigma). Fractionations were performed as described previously. Briefly, cell pellets from a 90% confluent 10-cm plate were gently scraped and collected in PBS at room temperature (4000 rpm for 1 min to pellet). Pellets were lysed in 200 µL of ice-cold cytoplasmic lysis buffer for 5 min on ice. Cellular lysates were passed through 500 µL of an ice-cold sucrose cushion at 10,000 rpm for 10 min at 4°C. The ∼700-µL cytoplasmic fraction was removed and used for protein analysis or further processed to obtain RNA samples. Pelleted nuclei were rinsed gently with 200 µL of ice-cold PBS and then resuspended in 100 µL of ice-cold glycerol buffer prior to being lysed with 100 µL of ice-cold nuclear lysis buffer. After vortexing twice for 2 sec, nuclear lysates were incubated on ice for 2 min and then pelleted at 14,000 rpm for 2 min at 4°C. The soluble nuclear fraction was removed and used for protein analysis or further processed to obtain RNA samples. The chromatin pellet was rinsed gently with 200 µL of ice-cold PBS and then digested in 50 µL of 1× DNase I buffer with 2 U of Turbo DNase (Thermo Fisher) for 10 min at 37°C. Protein samples were removed before dissolving the digested chromatin in 1 mL of TRIzol (Life Technologies) for RNA isolation. All centrifugations were conducted in Eppendorf 5424 centrifuges at either room temperature or 4°C. All solutions were supplemented with 1× Complete protease inhibitors (Roche Life Sciences). In total, we prepared three biological replicates of cell fractionation into chromatin, nucleoplasm, and cytoplasm. Both RNA-seq and m6A mapping analysis are highly reproducible among biological replicates, and Figure 1 integrates the result of three biological replicates.

51We also included in the Supplemental Material a protocol for m6A-CLIP with a detailed day-to-day arrangement of experiments.

52To rigorously classify m6A peaks into exonic versus intronic regions, we used the complete set of transcript annotations according to GENCODE (version 19 of human hg19). If an m6A peak was located at any known exonic region, it was classified as an exonic m6A peak. If it was located only at an intronic region of transcripts, it was classified as an intronic m6A peak. Using alternative annotations (e.g., Ensembl 85 version 25 for human GRCh38) generated essentially the same result. To examine the distribution of m6A peaks in partially spliced mRNAs (Fig. 1B,C), we focused on pre-mRNAs with abundant intronic RNA reads (i.e., there were two or more times more intronic RNA reads than exonic RNA reads). Using data for all pre-mRNAs, including those in which internal exons were completely or largely spliced (no or few introns), we obtained the same result: m6As in CA-RNAs are >93% in exons (see Fig. 1B).

53To unambiguously assign m6A peak regions to mRNAs according to gene annotations, we used a subset of GENCODE annotations (version 19 for human hg19 and version M10 for mouse mm10) by taking only one transcript isoform for each mRNA: We used the isoform with the longest mRNA length (alternatively, we used the isoform with the most distal 3′ end, which generated essentially the same result). We then removed overlapping transcripts from the set to avoid any ambiguity in determining which transcript the m6A peak region was from.

54We considered m6A peak regions that were within 1-kb mRNA distance from stop codons and the fact that mRNAs had different lengths. We generated 100 intervals, each with a 10-nt size for 1 kb upstream of and downstream from the stop codons. We computed the “m6A peak region density” (i.e., “m6A peak density”) for each interval as follows: We scanned through all mRNAs of interest that contained this interval and examined whether an m6A peak region existed in this interval. We then enumerated those cases that contained m6A peak regions and divided this value by the total number of mRNAs that contained this interval. For the plot of m6A peak region density around the start of last exons and start codons, we performed the same analysis except anchoring at the start of the last exons and start codons.

55To identify pre-mRNA reads that have both intron and m6A-containing exon sequences, we focused on precisely mapped m6A sites that were within 80 nt of splice sites (the length of our m6A sequencing reads was ∼80 nt). We required the read length in the intronic region to be at least 4 nt for these pre-mRNA reads. Furthermore, we also required the number of pre-mRNA reads in each case for CA-RNA be more than two as the reliable evidence of existence. In addition, these pre-mRNA reads should not exist in nucleoplasm and cytoplasm RNAs. We identified >200 internal exons to have m6A-containing exon–intron junction fragments (six total examples in Fig. 3B,C and Supplemental Fig. 5; full list in Supplemental Table 1).

56To clearly show the distribution of m6A peaks relative to splice sites, we focused on internal exons with exon length at least 200 nt so that the 100-nt exon regions from 5′ splice sites and 3′ splice sites would not overlap (Fig. 4A,C). The internal exons that are at least 200 nt long contained ∼80% of all internal exon m6As. “Relative m6A peak density” for a fixed position relative to a splice site (Fig. 4A,C) was calculated as the m6A peak density at that position scaled in proportion to the average m6A peak density in exonic regions at least 100 nt away from the splice sites. Figure 4B shows that ∼7% of exonic m6As are within 50 nt of splice sites for internal exons in Figure 4A (i.e., at least 200 nt long). If we consider all m6A-containing internal exons, including those exons >100 nt long in which all m6As are automatically within 50 nt distance from splice sites, 20% of exonic m6As are within 50 nt of splice sites.

57To determine m6A peaks that are higher in CA-RNA, for each m6A peak region, we enumerated reads of m6A immunoprecipitation and the input for CA-RNA and nucleoplasm RNA to evaluate the statistical significance (Fisher's exact test). Benjamini-Hochberg was implemented to adjust the P-value to the FDR for multiple testing. The requirement that an m6A peak region is higher in CA-RNA included (1) that the reads of mRNAs in m6A peak regions were adequate for m6A peak region detection in both CA-RNA and nucleoplasmic mRNA (reads per kilobase per million mapped reads [RPKM] ≥1) and (2) that the m6A peak regions that are higher in CA-RNA were determined by requiring FDR ≤0.05 and an at least twofold higher peak region enrichment in CA-RNA compared with nucleoplasmic mRNA. At a lower cutoff (e.g., ≥1.5 fold), the same conclusion held: Most m6A peaks are modified with the same level between CA-RNA and nucleoplasmic mRNA. Comparison of individual m6A peak signal strength in nucleoplasmic RNA and cytoplasmic RNA for the same m6A peak was performed in the same way as the comparison between CA-RNA and nucleoplasmic RNA.

58We determined the T1/2s of individual polyA+ mRNAs in mouse ESCs and Mettl3 knockout by sequencing after five time points (0, 1, 2, 4, and 8 h) of actinomycin (final concentration of 5 µg/mL; Sigma, no. A9415) treatment in three biological replicates. The T1/2 was determined as ln(2)/k, where k is the decay rate constant. The individual mRNA abundance levels at different time points after actinomycin D treatment were fitted to a first-order exponential decay curve to calculate the decay rate constant (k). A single T1/2 was calculated for each mRNA by using all three replicate values at each time point. For downstream detailed analysis, we considered only the T1/2s for individual mRNAs with reliable statistics (i.e., FDR <0.05, fitting the exponential decay model by t distribution). We used the published mRNA T1/2 data for HeLa cells to overlay HeLa cell m6A data with mRNA T1/2 (Fig. 5A). Using the mRNA T1/2 data in human hepatoma cancer cells from an independent group, we obtained similar associations of more m6As in mRNAs and shorter T1/2s (Supplemental Fig. 10).

59The m6A minigene for mRNA half-life validation was constructed based on a common retroviral GFP vector (Addgene, no. 1764); puromycin was the selection marker for successful DNA integration. The mRNA regions containing m6A sites were cloned in-frame into the minigene at multiple clone sites that were at the end of the coding region of GFP. Synonymous point mutations were carefully made to disrupt the m6A RAC core motif while not changing the underlying protein-coding sequence.

60The detailed sequences for the Ppp1r8 and Sox2 constructs (wild type and mutant) are in the Supplemental Material.

61Global quantification of m6A by mass spectrometry was practiced according to a protocol described previously. To determine m6A peak regions that were lost due to Mettl3 knockout in mouse ESCs, we practiced the analysis as reported previously. For each m6A peak region, we enumerated reads of m6A immunoprecipitation and the input for wild-type and knockout mouse ESCs to evaluate the statistical significance (Fisher's exact test). Benjamini-Hochberg was implemented to adjust the P-value to the FDR for multiple testing. The requirement that an m6A peak region is considered lost due to Mettl3 knockout demanded (1) that the mRNAs containing these m6A peak regions should be adequately expressed in both wild-type and knockout mouse ESCs (RPKM ≥1), (2) that the expression of mRNAs in m6A peak regions was adequate for m6A peak region detection in both knockout and wild type (RPKM ≥1), and (3) an FDR ≤0.05 and an at least twofold decrease of peak region enrichment in knockout compared with wild type.