Dynamic m⁶A modification regulates local translation of mRNA in axons

Abstract

N⁶-methyladenosine (m⁶A) is a reversible modification in mRNA and has been shown to regulate processing, translation and decay of mRNA. However, the roles of m⁶A modification in neuronal development are still not known. Here, we found that the m⁶A eraser FTO is enriched in axons and can be locally translated. Axon-specific inhibition of FTO by rhein, or compartmentalized siRNA knockdown of Fto in axons led to increases of m⁶A levels. GAP-43 mRNA is modified by m⁶A and is a substrate of FTO in axons. Loss-of-function of this non-nuclear pool of FTO resulted in increased m⁶A modification and decreased local translation of axonal GAP-43 mRNA, which eventually repressed axon elongation. Mutation of a predicted m⁶A site in GAP-43 mRNA eliminated its m⁶A modification and exempted regulation of its local translation by axonal FTO. This work showed an example of dynamic internal m⁶A demethylation of non-nuclear localized mRNA by the demethylase FTO. Regulation of m⁶A modification of axonal mRNA by axonal FTO might be a general mechanism to control their local translation in neuronal development.

INTRODUCTION

N⁶-methyladenosine (m⁶A) is the most widely distributed internal modification in mRNA (1–3). m⁶A modification of mRNA is a dynamic and reversible process which occurs in nuclear speckles where the methyltransferases (‘writers’) and demethylases (‘erasers’) are concentrated (4). The methyltransferase complex including METTLE3, METTL14 and WTAP mediates m⁶A methylation (5), whereas Fat mass and obesity-associated protein (FTO) (6) and ALKBH5 (7), two demethylases, catalyze demethylation of m⁶A. m⁶A modification has been shown to regulate pluripotency of embryonic stem cells and somatic cell reprogramming (8–11), sex determination (12,13), ultraviolet (UV)-induced DNA damage response (14), spermatogenesis (15), haematopoietic stem and progenitor cell specification (16), T cell homeostasis (17), myeloid differentiation (18), cortical neurogenesis (19) and neural circuitry (20). However, the function of m⁶A in neural development is still largely unknown. FTO which was originally associated with obesity (21), is the first identified m⁶A eraser (6). Previous studies on FTO in nervous system have focused on its roles in postnatal and elder brains including postnatal growth, brain volume, memory and adult neurogenesis (22–25). However, whether and how FTO regulates neuronal development such as axon growth remains unclear.

Messenger RNAs can be transported and targeted to different subcellular domains and then translated (26–28). Mammalian neurons are extremely polarized cells and mRNA can be transported and stored in their distal subcellular compartment such as axons (29). With the translational machinery in axons, axonal mRNA can be translated locally in axons in response to external cues (30,31). Local translation in axons provides an ideal and economical mechanism for axons (and their growth cones) to rapidly synthesize new proteins needed (32). Local translation in axons has been shown to regulate axon guidance (33–36), axon growth (37,38), axon injury (39), axon regeneration (40,41), neurodegeneration (42), neuronal survival (43), neuronal specification (44), formation and maintenance of neural circuits (45). However, whether ‘writing/erasing’ of m⁶A in mRNA can take place in axons and whether m⁶A mRNA modification can regulate local translation in axons are completely unknown.

Here, we investigated whether m⁶A mRNA modification can regulate local translation in axons. We found that FTO, the m⁶A eraser which is widely believed to demethylate m⁶A in nucleus, is enriched in axons. Surprisingly, we found that Fto mRNA exists in axons and can be locally translated. We confirmed that GAP-43 mRNA is m⁶A-modified. Our data demonstrated that specific inhibition or knockdown of axonal FTO led to elevation of m⁶A levels of axonal GAP-43 mRNA, suppression of local translation of GAP-43 and blocking of axon elongation. Furthermore, our experiments showed that this inhibition effects could be rescued by mutating m⁶A site of GAP-43 mRNA. These findings provide important insight into the functional role of m⁶A modification in regulating local translation in neural development.

MATERIALS AND METHODS

Neuronal culture

The use of animals and protocols were approved by the Laboratory Animal Welfare and Ethics Committee of Southern University of Science and Technology. All reagents used for neuronal cultures were from Thermo Fisher Scientific (USA) unless otherwise specified. Dorsal root ganglia (DRG) neurons were dissected from E13 C57BL/6 mouse embryos in Leibovitz’s L-15 medium and dissociated in TrypLE express enzyme. Dissociated neurons or explants of DRG were plated on acid-washed glass coverslips pre-coated with poly-D-lysine (100 μg/ml, Trevigen) and laminin (3.3 μg/ml, Trevigen), and cultured in Neurobasal medium supplemented with B27 (20 ml/l), GlutaMAX Supplement (2 mM), Penicillin–Streptomycin (100 U/ml, 100 μg/ml), 5-fluoro-2′-deoxyuridine (10 μM, Sigma) and Nerve growth factor (NGF) (50 ng/ml). For bath application of rhein, DRG neurons were plated and cultured for 2 h. Then rhein (Sigma) was added to the medium with final concentration of 50 μM. After 4 h, neurons were either processed for immunostaining or used for axon growth assay by taking phase-contrast images.

Knockdown or overexpression using lentiviral system

The lentiviral knockdown plasmid pLKO.1-eGFP was made from pLKO.1-TRC by replacing Puro^R with eGFP. The target sequences of shRNA are as following: shFto, 5′-GCATGTCAGACCTTCCTAAAG-3′; shAlkbh5, 5′-GCATACGGCCTCAGGACATTA-3′; shMettl3, 5′-CGTCAGTATCTTGGGCAAATT-3′; shMettl14, 5′-CCTGAGATTGGCAATATAGAA-3′; shYthdf1, 5′-GCACACAACCTCTATCTTTGA-3′; shYthdf2, 5′-GGACGTTCCCAATAGCCAACT-3′; shYthdf3, 5′-GCACCTAAACCAACTTCTTGG-3′; shLacZ (control), 5′-GCATAAACCCGCCACTCATCT-3′. The lentiviral overexpression vector pLVX- IRES-eGFP was made from pLVX-IRES-Hyg by replacing Hyg^R with eGFP. The lentiviral particles were generated using a second generation packaging system. In brief, HEK293T cells were transfected with 10 μg shRNA-expressing construct or Fto-expressing construct, and 8.9 μg pHR and 1.1 μg pVSVG in Opti-MEM (Thermo Fisher Scientific) containing Lipofectamine 3000 (Thermo Fisher Scientific). Cell culture medium containing virus was collected and filtered 48 h post-transfection and then subjected to ultracentrifugation at 22 000 rpm for 2 h at 22°C. The viral pellet was resuspended in 1% bovine serum albumin (BSA) in phosphate-buffered saline (PBS) and stored at −80°C. Neurons were analyzed by immunofluorescence (IF) 48 h after virus transfection.

Microfluidic chamber assays

The master mold for Poly(dimethylsiloxane) (PDMS) replication was fabricated using a two-layer SU-8™ (MicroChem) photolithography process as described in a previous publication (46). First, a silicon wafer with Al alignment marks was prepared. On a cleaned 3-inch silicon wafer, a thin layer of Al (150 nm) was deposited by an e-beam evaporator (HHV Ltd.) and alignment marks were defined by a conventional photolithography process using a positive photoresist (Rzj-304, Suzhou Ruihong Electronic Chemical Co.) followed by an Al etching (Al etchant Type-A, Transene) and stripping of the photoresist (Remover PG, MicroChem). On the silicon wafer with alignment marks, two layers of SU-8™ with different thicknesses were sequentially patterned. First, a 2.5 μm thick channel layer was patterned with SU-8™ 2002 and a 100 μm thick second layer was patterned on top of the first layer using SU-8™ 2075. Two layers were aligned and exposed to UV using a Karl Suss mask aligner (MA6, SÜSS MicroTec).

Microfluidic chambers were replicated from these master molds by cast molding PDMS pre-polymer (10:1 mixture, Sylgard^® 184, Dow Corning) at 80°C for 1 h. After punching reservoirs, both microfluidic chamber and cover glass surfaces were treated with corona discharge (BD-20AC, Electro-Technic Products) for 1 min. Then chambers were immediately assembled onto cover glasses. The devices were coated with poly-D-lysine and laminin subsequently. Dissociated DRG neurons were plated into cell body compartments. Axons will typically grow across the microgrooves to the axonal compartment by DIV2–3 for traditional bipartite chambers (Supplementary Figure S3B), and to the distal axon compartment by DIV3–4 for tripartite chambers (Supplementary Figure S3A).

For assays using severed axons, microfluidic chambers were opened by cutting cell body compartments before assembly. DRG explants were plated next to microgrooves and grown in microfluidic chambers for 2DIV and then the cell bodies were removed by forceps. By examining only severed axons, any possibility that changes of axonal proteins or m⁶A–mRNAs after treatments are due to contribution by soma is excluded.

Immunostaining

Cultured DRG neurons were rinsed with PBS and then fixed with 4% paraformaldehyde/PBS, pH 7.4. After fixation, the neuronal somas and axons were permeabilized and blocked with PBS/0.1% Triton X-100/1% BSA and then incubated with primary antibodies. Primary antibodies in this study were all well described and widely used in the field. The dilutions and sources of antibodies are as following: FTO (1:500, Phosphosolutions, 597-FTO), ALKBH5 (1:500, Sigma, HPA007196), METTL3 (1:1000, Proteintech, 15073–1-AP), METTL14 (1:1000, Sigma, HPA038002), YTHDF1 (1:500, Abcam, ab99080), YTHDF2 (1:1000, Proteintech, 24744–1-AP), YTHDF3 (1:500, Abcam, ab103328), m⁶A (1:1000, Synaptic Systems, 202003), Tau1 (1:1000, Millipore, MAB3420), Tau1 (1:1000, Abcam, ab76128) and GAP-43 (1:1000, Millipore, AB5220). Alexa Fluor-conjugated secondary antibodies (Thermo Fisher Scientific) were used at 1:1000. Fluorescent images were acquired using laser-scanning confocal microscopes (Leica SP8 or Nikon A1R) with LASX or NIS software, respectively. All images for each group in the same experiment were collected with identical settings. Quantification of IF signals in axons was performed using ImageJ software for each condition. Specific fluorescence signals from axons were normalized to axon area defined by Tau1 counterstains.

Axonal RNA RT-PCR

DRG neurons were grown in tripartite microfluidic chambers with two microgroove barriers (200 and 500 μm, respectively; illustrated in Supplementary Figure S3A). The design of the tripartite chambers makes sure that no neuronal soma will be present in the distal axon compartment. TRIzol reagent (Thermo Fisher Scientific) was applied to either soma or distal axon compartment (50 μl per compartment). Lysates from 10 chambers were pooled together and total RNA was extracted following the manufacturer’s protocol. After first-strand cDNA was synthesized using SuperScript IV First-Strand cDNA Synthesis System (Thermo Fisher Scientific), polymerase chain reaction (PCR) was performed with specific primers for mouse clones as following:

Fto, 5′-GAAGTGGTGAGGATCCAAGG-3′ and 5′-CTGCCTTCGAAGCTGGACTC-3′;

β-actin, 5′- AGGGAAATCGTGCGTGACAT-3′ and 5′-ACGCAGCTCAGTAACAGTCC-3′; Nrxn2, 5′- CAGACCTCATCGCTGACGC-3′ and 5′- AGGCGGTCCATCCGTGTAC-3′; H1f0, 5′-AGTATATCAAGAGCCACTACAAGG-3′ and 5′-AATGTATTTACAGAAAACAGGAGG-3′.

Axon growth assay

Bright field images were taken using an inverted microscope (Nikon LS100). Axons were manually traced and measured using Image Pro Plus software. For bath application, rhein (50 μM) were added to wells 2 h after plating cells. Three random fields were imaged before and 4 h after rhein treatment. For axon-specific treatment of rhein, DRG neurons were grown in microfluidic chambers and rhein was applied only to axons after axons grew to the axon compartment. The same axons were imaged and analyzed before and after rhein treatment.

Knockdown using siRNA

The designing and transfection of siRNAs were previously reported (44). The target sequences of siRNA are as following: siNC (RNAi negative control), 5′-UUCUCCGAACGUGUCACGUTT-3′; siFto-1, 5′-GUCAGACCUUCCUAAAGCUTT-3′; siFto-2, 5′-CAGGCACCUUGGAUUAUAUTT-3′; si5s/rRNA, 5′-CGUCUGAUCUCGGAAGCUATT-3′; si5.8s/rRNA, 5′-GCUACGCCUGUCUGAGCGUTT-3′; si18s/rRNA, 5′-GCUCGCUCCUCUCCUACUUTT-3′; si28s/rRNA, 5′-GCGCCUAGCAGCCGACUUATT-3′. The knockdown was validated by real-time reverse transcriptase (RT)-PCR with primers: Fto, 5′-GGAGAGGAAATCCATAATGAGGTG-3′ and 5′-CGCATTTGTCATGCTCTCCAT-3′; 5s rRNA, 5′-TCTACGGCCATACCACCCTGAACGC-3′ and 5′- GGTATTCCCAGGCGGTCTC-3′; 5.8s rRNA, 5′-GGTGGATCACTCGGCTCG-3′ and 5′- CAACCGACGCTCAGACAGG-3′; 18s rRNA, 5′ CTTTGGTCGCTCGCTCCT-3′ and 5′-AGGGAGCTCACCGGGTTG-3′; 28s rRNA, 5′-CGAGCTCAGGGAGGACAGAA-3′ and 5′-GCCGCCACAAGCCAGTTAT-3′. Measurement of GAP-43 mRNA levels after rhein treatment or siFto knockdown was performed by real-time PCR with primers: GAP-43, 5′-ATTCAGGCTAGCTTCCGTGG-3′ and 5′-GCATCGGTAGTAGCAGAGCC-3′.

siRNA-mediated compartmentalized knockdown of Fto mRNA or rRNA in axons were carried out using GeneSilencer siRNA Transfection Reagent (Genlantis) following manufacturer’s manual and protocols from previous studies (44,47). Briefly, siRNA was mixed with GeneSilencer Reagent and added directly to dishes (for bath application) or axonal compartment of microfluidic chambers (for compartmentalized knockdown). After 48 h, DRG axons and neurons were analyzed by IF or subjected to axon growth assay.

Anti-m⁶A immunoprecipitation

Immunoprecipitation (IP) of m⁶A-modified RNA using a specific m⁶A antibody (Synaptic Systems, 202003) was carried out following a published protocol (48). Briefly, total RNA (1 μg) was incubated with m⁶A antibody (16 μg/ml) in 1 × IP buffer (10 mM Tris–HCl, 150 mM NaCl, 0.1% Igepal CA-630, 0.4 U/μl RNasin, 2 mM Ribonucleoside vanadyl complexes) and then Dynabeads Protein G (Thermo) was added. After extensive washing with 1 × IP buffer, m⁶A-modified RNA was eluted using Elution buffer (1 × IP buffer, supplemented with 2.56 mg/ml m6A 5′monophosphate sodium). Eluted RNA was pelleted and used to synthesize cDNA. Then semi-quantitative RT-PCR was carried out to detect GAP-43 mRNA levels using primers GAP-43-p1, 5′-ATTCAGGCTAGCTTCCGTGG-3′ and 5′-GCATCGGTAGTAGCAGAGCC-3′; GAP-43-p2, 5′-GAGGGAGATGGCTCTGCTACTA-3′ and 5′-AGCCTCGGGGTCTTCTTTAC-3′.

Constructs and nucleofection

The complete cDNA with 3′ untranslated region (3′ UTR) encoding GAP-43 was obtained by RT-PCR from total RNA of DRG. The predicted m⁶A site (the A immediately following the stop codon: …TGA’A’…) was mutated to C (…TGA’C’…) by PCR. Either wild-type (WT) or m⁶A-mutated GAP-43 cDNA with 3′UTR was cloned to the vector pCAGGS-IRES-eGFP, generating expression constructs pCAGGS-GAP43^WT-IRES-eGFP and pCAGGS-GAP43^MT-m6A-IRES-eGFP, respectively. Nucleofection of the constructs was performed using a 4D-Nucleofector System machine (LONZA) and P3 Primary Cell 4D-Nucleofector X Kit (LONZA) following the manufacturer’s instructions. Briefly, dissociated DRG neurons were re-suspended in Nucleofector Solution (5 × 10⁵ − 10 × 10⁵ cells in 20 μl for each condition). After 1 μg of plasmid was added, the mixture was transferred to a well in the 16-well Nucleocuvette Strips and run program DC-100. Then the cells were transferred from the well and plated to microfluidic chambers for culturing. After axons grew to the axonal compartment, rhein was added specifically to axons. Only GFP⁺ axons were measured for GAP-43 IF and checked for axon growth rates.

Statistical analysis

No statistical methods were used to pre-determine sample sizes, but our sample sizes were similar to those generally applied in the field. Data distribution was assumed to be normal, but this was not formally tested. Samples were randomly allocated to different groups. Data collection and analysis were not performed in a blinded manner to the conditions of the experiments. No data points were excluded from the analyses. Data are mean ± s.e.m. All experiments were conducted at a minimum of three independent biological replicates in the lab. Graphs and statistical analysis were performed using GraphPad Prism 5 software and SPSS software. A P-value < 0.05 was considered as statistically significant. One-way analysis of variance (ANOVA) with Tukey’s post test was employed to the comparison of three or more groups. Unpaired, two-tailed t-test was performed for comparison of changes between two groups.

RESULTS

FTO is enriched in axons and derived from local translation

To explore whether m⁶A can regulate local translation in axons, we first tested whether m⁶A writers, erasers and readers are expressed in axons. Embryonic day 13 (E13) DRG were dissected, dissociated and cultured in vitro. The expression of m⁶A writers, erasers and readers were examined by immunostaining of DRG neurons with antibodies which are widely used to specifically detect FTO (49), ALKBH5 (7), METTL3 (6), METTL14 (49), YTHDF1 (50), YTHDF2 (49) and YTHDF3 (50). As shown in Figure 1A and B, in addition to soma (both nucleus and cytoplasm), FTO is also enriched in axons. This is surprising since FTO is traditionally thought to work in nucleus and its substrate is m⁶A in nuclear RNA (6). The localization of FTO in axons suggests that this non-nuclear pool of FTO might play roles in axons and axonal RNA might be its substrate as well. The expression of other m⁶A writers, erasers and readers in axons are generally very low compared with FTO (Supplementary Figure S1). The specificity of these antibodies used in IF was further verified by knockdown experiments, in which the IF signals of these m⁶A writers, erasers and readers were gone or significantly decreased after lentiviral shRNAs against Fto, Mettl3, Mettl14, Alkbh5, Ythdf1, Ythdf2 and Ythdf3 were transfected to DRG neurons (Supplementary Figure S2).

The enrichment of FTO in axons suggests that it might be locally translated. To test this possibility, we first investigated whether Fto mRNA is localized in axons or not. For local translation studies in axons, microfluidic chambers are powerful platforms which can realize fluidic isolation of axons from soma (51). We cultured DRG neurons in tripartite microfluidic chambers which were designed to prepare pure axonal materials (44). The design with two barriers of microgrooves ensures that there is no possibility of cell body migration to the distal axon compartment (Supplementary Figure S3A). RT-PCR using axonal and soma RNA showed that Fto mRNA was detected in both axons and soma (Figure 1C). Quantitative analysis for expression levels showed that axonal Fto mRNA level is higher than β-actin and similar as Nrxn2 mRNA (two positive controls for axonal mRNAs (45,52)) (Figure 1D). The absence of H1f0 mRNA (a histone transcript restricted to soma which worked as a negative control (53)) from axons confirmed that the axonal material was pure with no neuronal soma incorporation (Figure 1C). The localization of Fto mRNA in axons suggests that it might be locally translated. To further test this, we cultured DRG explants in traditional bipartite microfluidic chambers (Supplementary Figure S3B). After axons grew to axonal compartment, DRG culture medium was replaced with NGF-free medium for 4 h. After starvation of NGF, explants were removed to exclude the contribution of axonal FTO by soma and the severed axons were treated with NGF for 1 h. We found that NGF treatment led to increased FTO IF signals in axons, which could be blocked by co-application of cycloheximide, a protein synthesis inhibitor (Figure 1E and F). These data suggest that FTO could be locally translated in axons.

To further confirm that FTO is derived from local translation in axons, we performed compartmentalized knockdown in axons with siRNAs specifically against Fto mRNA. First, we tested the knockdown efficiency of siRNAs. Bath application of two siRNAs against FTO (siFto-1 and siFto-2) to cultured DRG neurons showed a significant knockdown (52 ± 2%, 45 ± 4%, respectively) of Fto mRNA compared with a negative control siRNA (siNC) using real-time PCR (Figure 1G). Then, we did compartmentalized knockdown in axons by applying siFto-1, siFto-2 to axonal compartment in microfluidic chambers. Compared with siNC, axon-specific knockdown resulted in 70 and 66% decrease of FTO IF signals in axons by siFto-1 and siFto-2, respectively (Figure 1H and I). Axonal knockdown of FTO did not change FTO protein levels in soma (Supplementary Figure S3C and D). These data suggest that axonal FTO is derived from local translation in axons.

Inhibition of FTO increases m⁶A signals in axons and represses axon elongation

Next, we wanted to explore the functions of axonal FTO in neurons. Rhein, a natural product, was discovered to be a potent small-molecule inhibitor of FTO (54). The good inhibitory activity of the cell-active rhein makes it a powerful tool to study FTO functions in neurons. In normal conditions, m⁶A levels in axons are much lower than soma in cultured DRG neurons (Figure 2A and B). Bath application of rhein led to a moderate increase (0.26-fold increase compared with vehicle treatment) of m⁶A in soma (Figure 2C and D). However, a dramatic increase (0.95-fold increase) of m⁶A signals in axonal RNAs by rhein treatment was observed (Figure 2E and F). The normally low level, and the significant upregulation of m⁶A levels in axonal RNAs by rhein treatment suggest that axonal RNAs are the preferred substrates of axonal FTO which keeps axonal RNAs in a de-(m⁶A)methylated state. Knockdown of axonal ribosomal RNAs (rRNA) did not change m⁶A levels in axons (Supplementary Figure S4A–F), suggesting that m⁶A IF signals and their changes by rhein treatment were mainly observed in axonal mRNAs. This is consistent with the findings that most modifications of rRNA are buried within the ribosome and are not easily accessible (3). We continued to check the effects of rhein treatment on neuronal development. Four hours treatment of rhein in cultured DRG neurons significantly inhibited axon elongation, with axon growth rate of 22.17 ± 1.74 μm/h for vehicle control reducing to 11.20 ± 0.66 μm/h for rhein treatment (Figure 2G and H). These results support that FTO is required for axon elongation.

Next, we wanted to inhibit FTO specifically in axons. DRG explants were grown in microfluidic chambers and after axons grew to the axonal compartment, explants were removed from axons to exclude any further contribution of m⁶A modified RNA from soma. Then severed axons were treated with rhein for 2 h. Compared with vehicle treatment, axonal m⁶A signals showed a significant increase after rhein treatment (Figure 3A and B). These data suggest that FTO in axons is required to regulate de-(m⁶A)methylation of RNAs in axons. We continued to test whether axon-specific treatment of rhein could inhibit axon elongation. Axon-specific treatment of rhein in microfluidic chambers for 4 h resulted in a significant decrease of axon elongation rate compared with vehicle control, with axon growth rates of 8.91 ± 1.15 and 23.08 ± 2.22 μm/h, respectively (Figure 3C and D). These results confirmed that activity of axonal FTO was required for axon elongation. To further confirm that the effects on m⁶A levels and axon elongation by rhein treatment were indeed mediated by FTO, we first prepared axonal-FTO-deficient axons (Figure 1H and I) and then we applied rhein to axons. Under these conditions, rhein treatment could not change m⁶A levels or axon elongation, suggesting rhein works through FTO in these assays (Supplementary Figure S5A and B).

To confirm more directly that FTO regulates m⁶A modification and axon elngation in axons, we carried out compartmentalized knockdown of Fto in axons using siRNAs (Figure 1H and I). After axon-specific knockdown of Fto, axonal m⁶A levels showed significant increases (Figure 3E and F). Overexpression of FTO led to dramatic decreases of m⁶A signals both in soma and axons (Supplementary Figure S5C–J). Axon-specific knockdown of Fto also significantly repressed axon elongation (Figure 3G). Together, these results suggest that axonal FTO is required for maintaining axon RNA in a low m⁶A modification state, and increased axonal m⁶A levels by inhibition of FTO (with rhein) or knockdown of Fto in axons repress axon elongation.

Axonal FTO regulates local translation of GAP-43 in axons

Local translation of mRNAs targeted to axons and growth cones can regulate neural development. Expression of the growth-associated protein-43 (GAP-43) is required for axon elongation (55). Targeting and local translation of GAP-43 mRNA in axons regulate axon elongation (37,56,57). However, the mechanisms regulating local translation of GAP-43 during axon elongation are still not clear. The increased m⁶A mRNA modification levels and decreased axon elongation by inhibition or knockdown of axonal FTO inspired us to hypothesize that m⁶A modification regulate axon elongation by controlling local translation of GAP-43.

To test this hypothesis, we wanted to know whether FTO can regulate GAP-43 protein levels in axons. Both bath application and axon-specific treatment of rhein led to dramatic decreases of GAP-43 protein level in axons (Figure 4A–D). FTO could control GAP-43 protein level through different mechanisms. First we examined whether FTO could control GAP-43 mRNA level. Neither rhein treatment nor knockdown of FTO could change GAP-43 mRNA levels (Figure 4E and F). These data suggest that FTO does not regulate transcription or stability of GAP-43 mRNA. Then we continued to test whether FTO could regulate local translation of GAP-43 mRNA in axons. Axon-specific co-application of rhein with MG-132, a specific and potent proteasome inhibitor, resulted in a significant decrease of GAP-43 protein in axons compared with MG-132 treatment alone (Figure 4G and H). Since MG-132 blocked protein degradation, any loss of axonal GAP-43 protein by rhein treatment could be only due to inhibition of local translation of GAP-43 mRNA. Overexpression of FTO led to significant increases of GAP-43 protein levels both in soma and axons (Supplementary Figure S6A–D).Together, these experiments confirm that axonal FTO could regulate local translation of GAP-43 in axons.

Axonal FTO de-(m⁶A)methylates GAP-43 mRNA in axons

Next, we explored the mechanisms by which axonal FTO regulates local translation of GAP-43 in axons. We hypothesized that axonal FTO regulate local translation of GAP-43 mRNA by controlling its m⁶A modification levels in axons. First, we wanted to know whether GAP-43 mRNA is modified by m⁶A. Analysis of GAP-43 mRNA using a mammalian m⁶A site predictor named SRAMP (sequence-based RNA adenosine methylation site predictor) (58) demonstrated that GAP-43 mRNA is modified by m⁶A and the m⁶A site is the A immediately after the stop codon (TGA’A’). To verify this, we carried out immunoprecipitation (IP) assay using total RNA extracted from E13 mouse DRG explants and a widely used m⁶A antibody (1,2,6,8,9,20). Compared with control IgG, m⁶A antibody pulled down GAP-43 mRNA which was detected by RT-PCR (Figure 5A). Interestingly, IP experiment done with RNA from E13 mouse spinal cord had similar results (Figure 5B). This raises an intriguing possibility that m⁶A modification of GAP-43 may be a recurring feature in nervous system. To confirm the m⁶A site in GAP-43 mRNA, we generated an expression construct of GAP-43 with its 3′UTR, in which the predicted m⁶A site (TGA’A’) was mutated to TGA’C’. Overexpression constructs of WT GAP-43 (‘WT’) and m⁶A-mutated GAP-43 (‘MT^m6A’) were transfected to DRG neurons by nucleofection and total RNAs were extracted respectively after rhein treatment. After IP with m⁶A antibody, much less GAP-43 mRNA with m⁶A site mutated (‘MT^m6A’) was pulled down compared with ‘WT’ control (Figure 5C; Supplementary Figure S6E). Together, these data suggest that GAP-43 mRNA is modified by m⁶A at the A after stop codon.

As experiments shown earlier, axonal FTO keeps axonal RNAs in a de-(m⁶A)methylated state (Figure 2A, B, E and F). We hypothesized that GAP-43 mRNA is modified by m⁶A in soma, and once transported to axons, GAP-43 mRNA is de-(m⁶A)methylated by axonal FTO and locally translated to promote axon elongation. This hypothesis is supported by the results that axon-specific loss of function of FTO by rhein treatment inhibited local translation of GAP-43 (Figure 4C, D, G and H) and axon elongation (Figure 3C and D). To further test this hypothesis, we carried out experiments to examine whether rhein treatment could increase m⁶A modification levels of GAP-43 mRNA. Indeed, rhein treatment led to significant increases of m⁶A levels of GAP-43 mRNA, which was demonstrated by more GAP-43 pulled down by m⁶A antibody after rhein treatment compared with vehicle control (Figure 5D).

Taken together, all the above data support that GAP-43 mRNA is modified by m⁶A, and in axons it is de-(m⁶A)methylated by axonal FTO in order to be locally translated and promote axon elongation.

Mutation of m⁶A site in GAP-43 mRNA can rescue inhibition of GAP-43 local translation and axon elongation by rhein

We next wanted to test whether mutation of m⁶A site in GAP-43 mRNA can rescue the inhibition of its local translation by rhein. Overexpression constructs of WT GAP-43 (‘WT’) and m⁶A-mutated GAP-43 (‘MT^m6A’) were transfected to DRG neurons by nucleofection and axonal GAP-43 protein levels were measured by IF. Axon-specific rhein treatment inhibited axonal GAP-43 protein levels compared with vehicle control with overexpression of WT GAP-43 (Figure 5E). This is similar with results by measuring endogenous axonal GAP-43 protein levels after rhein treatment in axons (Figure 4C and D). These results suggest that failure to de-(m⁶A)methylate axonal GAP-43 mRNA due to rhein treatment, which kept axonal GAP-43 mRNA in a m⁶A-modifed state, inhibited local translation of GAP-43 in axons. However, GAP-43 protein level with overexpression of the m⁶A-mutated GAP-43 (‘MT^m6A’) was not subject to rhein inhibition (Figure 5E), because GAP-43 mRNA was now in a de-(m⁶A)methylated state and could be locally translated. Similar results were observed with axon growth rate (Figure 5F). These results further support that axonal GAP-43 mRNA is de-(m⁶A)methylated by axonal FTO and then locally translated.

DISCUSSION

FTO is traditionally thought to work in nucleus (6). Here we find a novel and unexpected expression and localization of FTO in axons. Another surprising finding is that these non-nuclear FTO is derived from local translation of Fto mRNA in axons (illustrated in Figure 6A). Previously it was thought that substrate of FTO is m⁶A in nuclear RNA (6). Here we find that these axonally derived FTO can de-(m⁶A)methylate m⁶A in axonal RNA such as GAP-43 mRNA, which can then be locally translated in axons to promote axon elongation. Axon-specific loss-of-function of FTO by axonal rhein treatment or compartmentalized siRNA knockdown of GAP-43 in axons inhibited de-(m⁶A)methylation, which resulted in maintaining m⁶A modification of axonal GAP-43 mRNA (Figure 6B). These GAP-43 mRNA with maintained m⁶A modification cannot be locally translated, which results in axon growth inhibition (Figure 6B). Thus FTO has a surprising localization site–axon, previously unknown substrates–axonal RNA and a novel function–regulating local translation in axons.

It is known that m⁶A modification is reversible and dynamic (59,60). But whether there is a constitutive state of demethylation of m⁶A and whether this demethylation has a biological role have been under debate. Here we provide an example that dynamic m⁶A demethylation of axonal mRNA (catalyzed by axonally derived FTO) does occur in axons and has significant roles, e.g. regulating local translation in axons and axon elongation. Our study meets the call for evidence of dynamic cycle of m⁶A writing/erasing in biological systems (12,61).

Local translation of mRNA in axons plays important roles in neural development including axon guidance, growth and neuronal specification. Local translation can be regulated by various cis-elements in axonal mRNAs. For example, some cis-acting motifs in axonal mRNAs are required for axonal transport and targeting of mRNA (62,63). miRNA target sequences in axonal mRNA can recruit miRNA such as miR-132, miR-181d and miR-182, and regulate local translation (64–66). There are also other specific motifs in axonal mRNA which can interact with RNA-binding proteins such as ZBP1 and Fus, and regulate local translation (35,67,68). Here, we report a novel regulatory mechanism for local translation, the internal m⁶A modification of mRNA. Axonal GAP-43 mRNA which maintains m⁶A-modification due to loss-of-function of axonal m⁶A eraser FTO cannot be locally translated. These findings support a model that m⁶A-modified mRNAs are transported to axons in a silent (not translated) form, and they can be locally translated only when they are de-(m⁶A)methylated by axonal FTO.

GAP-43 mRNA with maintained m⁶A-modification fails to be locally translated, which is not due to RNA degradation because GAP-43 mRNA levels are not changed by rhein or knockdown of FTO. Inhibition of translation (without affecting mRNA stability) by maintained or excess m⁶A modification is consistent with findings that highly methylated mRNA showed significantly lower translation (69), and too much m⁶A modification has a negative effect on translation (70).

Our findings are consistent with previous reports that m⁶A is the target of FTO. A recent study suggests that FTO demethylates m⁶A_m (N⁶, 2′-O-dimethyladenosine) in the 5′ cap of mRNA more efficiently than the internal m⁶A (71). Our results suggest that m⁶A in axonal GAP-43 mRNA is indeed the target of axonal FTO. Mutation of the predicted m⁶A site (TGA’A’) in GAP-43 mRNA eliminates its modification, and exempts regulation of its local translation by axonal FTO. These results, together with our findings that inhibition of FTO has much more pronounced effects on axonal m⁶A than soma m⁶A, suggest that the preferred targets of axonal FTO are axonal m⁶A. Demethylation of m⁶A_m by FTO reduces the stability of m⁶A_m mRNAs (71). However, we find that de-(m⁶A)methylation of axonal GAP-43 mRNA by axonal FTO does not control its stability since rhein treatment or siFto knockdown does not change GAP-43 mRNA levels. Instead, axonal FTO can regulate local translation of GAP-43 in axons. Together, our results reveal a novel target (axonal m⁶A) for an unexpected pool of FTO (axonal FTO). Axonal transcriptome analysis has revealed mRNA encoding many proteins (29). It would be interesting to identify substrate mRNAs of axonal FTO in a transcriptome level. Regulation of local translation of mRNA by de-(m⁶A)methylation in axons might be a general mechanism in nervous system development.

SUPPLEMENTARY DATA

Supplementary Data are available at NAR Online.

ACKNOWLEDGEMENTS

We thank members of the Ji lab for technical support and helpful discussions; Caiguang Yang for reagents; Hongwei Guo and Samie R. Jaffrey for comments and suggestions on the manuscript.

FUNDING

Natural Science Fund of Guangdong Province [2016A030313638 to S.-J.J.]; Basic Research Grant from Science and Technology Innovation Commission of Shenzhen Municipal Government [JCYJ20160331115633182 to S.-J.J.]; China Thousand Talent Program for Young Outstanding Scientists Scholarship (to S.-J.J.); Startup Funds from Southern University of Science and Technology and Peacock Plan of Shenzhen Municipal Government [to S.-J.J. and J.P.]. Funding for open access charge: Startup Funds from Southern University of Science and Technology.

Conflict of interest statement. None declared.

REFERENCES

Author notes