FTO belongs to the non-heme Fe^II/α-KG-dependent dioxygenase AlkB family proteins that also includes ABH1 to ABH8^7,8. Among them, ABH1, ABH2, and ABH3 have been shown to oxidatively demethylate N-methylated DNA/RNA bases^9,10, and ABH8 catalyzes the hydroxylation of hypermodified tRNA wobble uridine^11,12. FTO has previously been shown to oxidatively demethylate m³T and m³U in single-stranded DNA (ssDNA) and single-stranded RNA (ssRNA) in vitro^7,13, although the observed activity is exceedingly low compared to those of the other AlkB family proteins¹⁴. The recent crystal structure of FTO confirms the substrate preference of FTO for single-stranded nucleic acids¹⁵. Therefore, modified RNAs are plausible substrate candidates for FTO.

Since the physiological substrates of the AlkB family proteins might not be limited to N¹-or N³-modified purines or pyrimidines, as in the case of ABH8^11,12, we turned our attention to N⁶-methyladenosine (m⁶A) (Fig. 1a), which is the dominating methylated base in mRNA with a frequency of 3–5 m⁶A per mRNA molecule in mammalian cells^16–19. Although the exact role of m⁶A in mRNA is largely unknown, it has been proposed to affect mRNA processing and export from nucleus to cytoplasm¹⁹.

We synthesized ssRNA and ssDNA with site-specifically incorporated m⁶A (Supplementary Scheme 1 and Supplementary Fig. 1). After treating an 8-mer ssDNA containing m⁶A (1 nmol) with an equal amount of recombinant human FTO (Supplementary Fig. 2) at pH 7.0 at room temperature overnight, we observed a loss of 14 Dalton in mass indicating that a demethylation process had occurred (Supplementary Fig. 3). To further investigate the demethylation activity of FTO towards m⁶A in DNA and RNA, we synthesized a series of 15-mer ssDNA and ssRNA containing a central m⁶A base. These single-stranded oligonucleotides were treated with 20 mol% of FTO at pH 7.0 at room temperature for 3 h. We then digested the oligonucleotides into nucleosides using nuclease P1 and alkaline phosphatase, and separated the products on a C18 column by HPLC. The HPLC trace profile indicated that m⁶A in ssDNA and ssRNA was completely converted to adenosine upon FTO treatment (Fig. 1b,c). We further tested a short m⁶A-containing RNA using a known sequence of cellular bovine prolactin (bPRL) mRNA, in which the position of m⁶A had been mapped^18,19. While this short RNA sequence can form a stem loop secondary structure, we still observed 40% of demethylation of m⁶A under the same conditions (Fig. 1d). We tested m⁶A-containing dsDNA and dsRNA with negligible activity observed under the same reaction conditions (0.2 equivalent of FTO for 3 h). After treatment with one molar equivalent of FTO at 16 °C overnight, we observed a demethylation yield of 40% for dsDNA and 24% for dsRNA, respectively (Supplementary Fig. 4).

To confirm that the observed demethylation requires the oxidative function of FTO, we tested m⁶A demethylation using several FTO mutants with active site mutations. Arg316 is a key residue for stabilizing the binding of α-KG. We expressed and purified the FTO R316Q mutant protein (Supplementary Fig. 2), and found that this R316Q mutation abolished 80% of the wild-type activity towards m⁶A demethylation in vitro (Supplementary Fig. 5). We subsequently constructed and tested two double mutant FTO proteins, H231A/D233A, in which two iron(II) ligands were mutated, and R316Q/R322Q, in which two α-KG ligands were mutated (Supplementary Fig. 2). Both of these double mutants completely lost m⁶A demethylation activity (Supplementary Fig. 5). These results confirmed that FTO catalyzes oxidative demethylation of m⁶A in an iron(II)- and α-KG-dependent manner.

We determined detailed reaction kinetics of FTO on a 15-mer m⁶A-containing ssRNA at pH 7.0 at 20 °C (Supplementary Fig. 6). The results are summarized in Supplementary Table 1 in comparison to our previously published kinetic data on m³U in ssRNA at pH 6.0. FTO exhibits a pH-dependent oxidative demethylation activity with the highest activity observed at pH 6.0¹³. Even if we compare the FTO activity towards m⁶A at pH 7.0 to that of m³U at pH 6.0 in ssRNA, we still observe an over 50-fold preference of FTO for m⁶A. To further validate this conclusion, we synthesized and tested ssRNA containing m³U with the same sequence as that used for m⁶A at 20 °C (Supplementary Fig. 7). The demethylation activity of FTO towards m⁶A is indeed significantly faster than towards m³U at neutral pH in vitro. Under physiological conditions, m⁶A is the best substrate discovered so far for FTO.

While m⁶A has not been identified in genomic DNA in higher eukaryotes, it is present in easily detectable levels in cellular mRNAs isolated from all higher eukaryotes and viral RNA^16–19. We explored whether m⁶A in mRNA is a physiological substrate of FTO in vivo. We transfected HeLa and 293FT cells with FTO siRNA (see Supplementary Methods), and western blotting of the total cell lysates from transfected cells showed close to 90% knockdown of FTO 48 h post-transfection (Supplementary Fig. 8a). We then isolated total mRNA from cells treated with control siRNA and FTO siRNA using a biotin-poly (dT) probe for separation followed by a rRNA depletion step to ensure depletion of rRNA (see Supplementary Methods and Supplementary Fig. 9). We then digested the total mRNA into nucleosides, and measured the relative amount of m⁶A levels by LC-MS/MS. The total contents of m⁶A and A were quantified based on a standard curve generated using pure standards, from which the m⁶A/A ratio was calculated (see Supplementary Methods and Supplementary Fig. 10). Our results indicated that knockdown of the cellular FTO increased m⁶A in mRNA by 23% in HeLa cells and 42% in 293FT cells (Fig. 2a). We also overexpressed wild-type FTO using a mammalian expression vector. Western blotting of the total cell lysates from transfected cells confirmed the overexpression of FTO by 6–8 folds after 24 h (Supplementary Fig. 8b). Total mRNA isolated from HeLa cells overexpressing the wild-type FTO showed a notable decrease of m⁶A (~18%) (Fig. 2b).

Since a large portion of m⁶A is after G in the consensus sequence (Pu[G>A]m⁶AC[A/C/U])¹⁹, we also employed a two-dimensional thin layer chromatography (2D-TLC) method that can detect the m⁶A content after G in mRNA (see Supplementary Methods). The results supported the conclusion that the m⁶A content in mRNA is affected by the cellular activity of FTO (Supplementary Fig. 11). A dot blot assay using an anti-m⁶A antibody was also used to qualitatively monitor the change of m⁶A level with the same conclusion obtained (Supplementary Fig. 12). We performed additional western blotting experiments for MT-A70 (AdoMet-binding subunit of mRNA N⁶-adenosine-methyltransferase) to confirm that the observed change of the m⁶A level was caused by FTO, but not from the potential alteration of the m⁶A mRNA methyltransferase expression²⁰ (Supplementary Fig. 13). Knockdown of FTO in HeLa cells led to a decrease of MT-A70 expression of around 15%, and overexpression of FTO led to an increase of MT-A70 expression of about 15%. Furthermore, knockdown of ABH3 in HeLa cells did not noticeably affect the total level of m⁶A in mRNA as a negative control (Supplementary Fig. 14). We conclude that the increase of the m⁶A level in the FTO knockdown sample is not due to an increased level of methyltransferase, but is a result of the depletion of cellular FTO.

Lastly, indirect immunofluorescence analysis was performed to visualize the localization of the FTO protein in mammalian cells. FTO is known to be entirely localized to cell nuclei⁷. We observed that the cellular FTO protein is present in a dot-like manner in nucleoplasm, and partially colocalizes with splicing or splicing-related speckle factors SART1(U4/U6.U5 tri-snRNP-associated protein 1) and SC35 (serine/arginine-rich splicing factor 2), and RNA polymerase II phosphorylated at Ser2 (Pol II-S2P) (Fig. 3a and Supplementary Fig. 15a,b), but does not colocalize with telomere marker TRF1, replication site PCNA, Cajal body marker Coilin, cleavage body marker CstF64, or P-body marker DCP1A (Supplementary Fig. 16a and 17). To investigate whether FTO is related to mRNA processing, we performed the same immunostaining experiment upon inhibition of Pol II-S2P transcription with actinomycin D (ActD). Quantitative foci analysis showed that the FTO foci number per cell was significantly decreased from 43±7 (−ActD) to 22 ± 5 (+ActD) with P-value less than 0.0001, similar to that observed for Pol–II S2P from 90 ± 12 (−ActD) to 24 ± 5 (+ActD), but not SC35²¹ (Fig. 3b), upon transcription inhibition (Supplementary Fig. 18). Intriguingly, the decreased FTO nucleoplasm foci became more concentrated into discrete speckle foci after transcription inhibition by ActD (Fig. 3a and Supplementary Fig. 15a), resembling that of Pol II-S2P²¹ (Supplementary Fig. 15a). Moreover, the transcription inhibition induced a more pronounced colocalization of FTO with both SC35 and Pol II-S2P (Fig. 3a and Supplementary Fig. 15a). Splicing speckles are major nuclear domains enriched for components of the splicing machinery such as SC35, polyA+ RNA, and numerous mRNA metabolic factors^21–24; genes associated with splicing speckles have been shown to function in spliceosome assembly or post-transcriptional splicing of pre-mRNAs^23,24. Knockdown of FTO does not affect the assembly of spliceosome (Supplementary Fig. 16b,c); however, FTO shows a distribution pattern in nuclear speckles similar to other splicing factors. FTO may be recruited to the speckles by its interacting partners in the assembled nuclear speckles. In conclusion, the partial colocalization of FTO with nuclear speckles further supports the concept that m⁶A in nuclear RNA is a substrate of FTO, and the enzymatic alteration may be linked to processing of recently transcribed mRNA.

In summary, we show that FTO efficiently demethylates m⁶A at neutral pH in vitro, and the level of m⁶A in cellular mRNA is affected by the oxidation activity of FTO in vivo. FTO partially localizes with nuclear speckles. The pronounced concentration of FTO in speckles upon Pol II-S2P transcription inhibition suggests a dynamic interaction of FTO with the nuclear speckles. In fact, MT-A70, a critical subunit of m⁶A-methyltransferase that introduces m⁶A into mRNA, has also been shown to localize in nuclear speckles, and probably associates with nuclear pre-mRNA splicing components²⁰. All these observations indicate that m⁶A in nuclear RNA is the physiological substrate of FTO, and that the function of FTO likely affects the processing of pre-mRNA and/or other nuclear RNAs.

Despite past research on m⁶A in mRNA, the biological function of this ubiquitous post-transcriptional modification remains unclear, particularly in mRNA from higher eukaryotes. Our cellular data suggest that a portion or subclass of m⁶A in mRNAs appeared to be affected by the activity of FTO in vivo. How the sequence and methylation status of m⁶A in mRNA affect the output of RNA still requires detailed investigation; the fact that the function of the major obesity factor FTO is to demethylate m⁶A in mRNA clearly indicates a novel, reversible regulatory mechanism present in mammalian cells. Methylation of DNA and histones contributes largely to epigenetic regulation in mammalian cells. The discovery of the FTO-mediated oxidative demethylation of m⁶A in nuclear RNA may initiate further investigations on biological regulation based on reversible chemical modification of RNA²⁵.

1