Cells and tissues

Human HeLa (cervical adenocarcinoma), HepG2 (hepatocellular carcinoma) and HEK293 (embryonic kidney) cell lines (ATCC) and primary mouse embryonic fibroblasts (MEFs) (C57BL/6, ATCC) were maintained in DMEM (Thermo Fisher Scientific) containing 25 mM glucose, 4 mM L-glutamine, supplemented with 100 U ml−1 penicillin, 100 μg ml−1 streptomycin and 10% FBS. Mouse embryonic stem cells (mESCs) were maintained in FBS-free N2B27-based media, as previously described43. Cells were routinely checked for mycoplasma contamination. Mouse tissues were obtained from wild-type and ob/ob C57BL/6 mice. For metabolic labelling, methionine-devoid DMEM was supplemented with L-methionine-(methyl-d 3 ) (Sigma-Aldrich).

Cell treatments

HepG2 cells were maintained as described above and subjected to the following growth conditions: (1) heat shock (43 °C for 4 h); (2) glucose starvation (glucose-depleted medium for 4 h); (3) amino acid starvation (amino acid-depleted medium, with or without 200 μM S-adenosyl-methionine (SAM) supplementation, for 4 h); and (4) serum starvation (serum-depleted medium for 4 h), following which they were immediately harvested in Cell Lysis Solution (5 Prime).

Yeast strains and growth conditions

Wild-type (BY4741) S. cerevisiae cells were grown vegetatively in YPD medium to mid-log phase and harvested. Wild-type heterothallic (Sp1) S. pombe cells were grown vegetatively in YES medium to mid-log phase and harvested. For induction of sporulation/arrest, Sp1 cells were washed and transferred to Edinburgh Minimal Medium lacking NH 4 Cl (EMM-N) for 4 h and harvested.

RNA purification

Total RNA from cells in culture and mouse tissues was purified using PerfectPure RNA Cultured Cell Kit (5 Prime) and PerfectPure RNA Tissue Kit (5 Prime), respectively, and DNase-treated. Enrichment of polyadenylated RNA (polyA + RNA) from total RNA was carried out using one round of GenElute mRNA Miniprep Kit (Sigma-Aldrich). Ribo-Zero Gold rRNA Removal Kit (Illumina) was used to deplete rRNA from polyA+ RNA before LC-MS/MS.

Dot blot assays

RNA Oligonucleotides were synthesized in-house with either m1A, m6A or A at a single internal position (5′-AC(m1A/m6A/A)UG-3′), spotted onto a nylon membrane (GE Healthcare) in decreasing amounts (1,000, 200, 40 and 8 pmol) and UV-crosslinked. Membranes were blocked with 5% non-fat dry milk in 1 × PBST (blocking buffer) for 1 h at 25 °C, and incubated overnight with either mouse anti-m1A antibody (1 μg ml−1, MBL) or rabbit anti-m6A antibody (1 μg ml−1, Synaptic Systems) in 1 × PBST at 4 °C. Following extensive washing with 1 × PBST, membranes were incubated with either HRP-conjugated goat anti-mouse IgG or anti-rabbit IgG antibody (1:2,500, Thermo Fisher Scientific) in blocking buffer for 1 h at 25 °C. Membranes were washed in 1 × PBST, developed with ECL substrate (Thermo Fisher Scientific) and imaged with FluorChem imager (Protein Simple) or X-ray film. Competitive dot blots were performed on separate membranes spotted with 75 pmol of the m1A-containing oligonucleotide by co-incubation of anti-m1A antibody with increasing concentrations of either m1A or m6A competitor mononucleoside (0, 1, 2 and 4 μM).

m1A detection and quantitation

Purified RNA (see ‘RNA purification’ section above) was subjected to liquid chromatography-tandem mass spectrometry (LC-MS/MS) for detection and accurate quantitation of m1A, essentially as previously described44. 200–400 ng of purified RNA was digested by P1 nuclease (Wako, 2 U) in 40 μl of buffer containing 25 mM NaCl and 2.5 mM ZnCl 2 for 2 h at 37 °C. Subsequently, 5 units (1 μl) of Antarctic Phosphatase (New England Biolabs) and 1 × Antarctic Phosphatase reaction buffer were added and the sample was incubated for another 2 h at 37 °C. The sample was then filtered (0.22 μm, Millipore) and injected into a C18 reverse phase column coupled on-line to Agilent 6410 QQQ triple-quadrupole LC mass spectrometer in positive electrospray ionization mode. Quantitation was performed based on nucleoside-to-base ion transitions (268-to-136 for A; 282-to-150 for m6A and m1A (retention times of 2.5 and 0.9 min, respectively); 285-to-153 for d 3 -m1A and d 3 -m6A; 245-to-179 for Ψ; 245-to-113 for U) using standard curves of pure nucleosides and stable isotope-labelled internal standards (d 3 -m1A and d 3 -m6A), as previously described45,46,47.

Synthesis of stable isotope-labelled internal standards

d 3 -m1A was synthesized following a previously described procedure for the synthesis of m1A48, replacing CH 3 I with CD 3 I. d 3 -m6A was synthesized following a previously described procedure for the synthesis of m6A49,50, replacing CH 3 NH 2 with CD 3 NH 2 .

m1A-seq

Mapping of m1A in total or mRNA was performed using m1A-seq, which is based on the previously described m6A-seq protocol51 with the following modifications: RNA fragmentation was performed using RNA Fragmentation Reagents (Thermo Fisher Scientific) for 15 min at 70 °C to minimize m1A-to-m6A rearrangement followed by column purification (RNA Clean & Concentrator, Zymo). Anti-m1A antibody (MBL) was pre-coupled to Protein G Dynabeads (Thermo Fisher Scientific) and used to immunoprecipitate methylated RNA fragments for 3 h at 4 °C. Fragment elution was carried out by either digestion with Proteinase K (Sigma-Aldrich; 5 units in 5 mM Tris pH 7.5, 1 mM EDTA and 0.25% SDS for 1.5 h at 37 °C) followed by TRIzol (Thermo Fisher Scientific) extraction of the supernatant and ethanol precipitation, or by competitive elution using m1A mononucleoside (Santa Cruz Biotechnology), as in m6A-seq. Induced m1A-to-m6A rearrangement was achieved by incubating the input and immunoprecipitation fragments in an alkaline buffer (50 mM Na 2 CO 3 , 2 mM EDTA, pH 10.4) for 1 h at 50 °C or 60 °C followed by column purification (RNA Clean & Concentrator, Zymo) (Extended Data Fig. 1m). Immunoprecipitated RNA fragments and comparable amounts of input were subjected to first-strand cDNA synthesis using the NEBNext Ultra RNA Library Prep Kit for Illumina (New England Biolabs). Sequencing was carried out on Illumina HiSeq2500 according to the manufacturer’s instructions, using 10 pM template per sample for cluster generation, TruSeq SR Cluster kit v3 (Illumina), TruSeq SBS Kit v3-HS (Illumina) and TruSeq Multiplex Sequencing primer kit (Illumina). Summary of read numbers for each replicate can be found in Supplementary Table 4.

Validation of m1A-seq

Reads were aligned to the 28S ribosomal RNA sequences of human (RefSeq NR_003287.2), mouse (RefSeq NR_003279/NCBI X00525) and S. cerevisiae (SGD RDN25-2, ID: S000006485.) using Bowtie252 with local alignment option (--local). Enrichment was calculated as the coverage ratio of immunoprecipitation to input. Mismatch rate was assessed using mpileup tool of the SAMtools software package (version 0.1.18)53 for calling variants over the entire 28S transcript. The frequency of non-adenosine bases at each adenosine position and the coverage of each base in the transcript were reported.

RNA expression level

Reads per kilobase of transcript per million mapped reads (RPKM) expression levels of RefSeq genes were calculated using HTseq-count54 and R package edgeR55. Fragments per kilobase of transcript per million mapped reads (FPKM) values were calculated by the CUFFLINKS tool (version 2.2.1)56. Only genes whose expression level was above the first quartile were considered adequately expressed and used for downstream analyses. Adequately expressed genes (RPKM) were divided into 10 expression bins for correlation with methylation and the fraction of methylated genes in each bin was calculated.

Peak calling

Adaptors and low quality bases were trimmed from raw sequencing reads using cutadapt57. Reads were aligned to the relevant genome (human-hg19, mouse-mm10, S. pombe- ASM294v2.29, and S. cerevisiae- S288C with UTR data downloaded from http://genie.weizmann.ac.il/pubs/PARS10/pars10_catalogs.html) using Tophat2 (version 2.0.12)58. Peaks enriched in immunoprecipitation over input experiments were identified using MACS2 (version 2.1.0.20140616)18. MACS2-identified peaks were intersected with a database of exons of the relevant genome (RefSeq annotation). Peaks were allocated to the feature containing the segment with which they share the largest overlap. Peaks falling in intergenic sequences or having an overlap shorter than 50 nucleotides were excluded from further analyses. For each cell type only peaks identified (FC ≥ 2 or FC ≥ 4, as indicated, FDR ≤ 0.05) in replicates were considered. Common human peaks were defined as peaks independently identified in both HeLa and HepG2 RNA. Negative peaks were identified by switching the immunoprecipitation and input samples.

Coverage analysis

The coverage of unique reads at each nucleotide position at the transcriptomic level, in immunoprecipitation and input reads, was calculated and reported. Only transcripts in which at least one nucleotide exceeded a minimum depth of 10 reads in immunoprecipitation were used in further analysis for reduction of background noise. For whole-transcript coverage plots, each transcript was divided into 100 bins of equal length, and the median coverage for each bin across all transcripts was calculated.

m1A stoichiometry

HepG2 mRNA was subjected to immunoprecipitation with anti-m1A antibody in 2 biological replicates. Input and immunodepleted (unbound sup) RNA samples (75 ng each) were hybridized to PrimeView human gene expression microarrays (Affymetrix) and expression levels were determined. Genes expressed below the first quartile (as determined by mRNA-seq, see above) were set as the minimal intensity limit (blank), and only genes with intensities above this threshold were considered. To correct for technical loss of RNA during the different steps of the procedure, the average ratio of sup/input intensity was calculated for never-methylated genes (considering only genes that were unmethylated in all m1A-seq experiments) and used for ratio correction of methylated genes carrying one m1A peak. Fractional methylation level per gene was calculated as 1 − (corrected sup/input ratio).

Identification of m1A-induced mismatches in m1A peaks

Identification of sequence variants in m1A peaks due to misincorporation at the m1A position during reverse transcription was carried out by determining the base composition at each position within peak regions using bam-readcount (https://github.com/genome/bam-readcount). Identified adenosine positions (according to the encoding DNA strand) were then filtered to exclude known genomic polymorphism sites (dbSNP version 141) and identified A-to-I editing sites (Radar database). Variants were considered if the mismatch rate was greater than 0.1 and the overall coverage was greater than 20 reads. For HepG2 samples the fold enrichment of mismatch rate in untreated over rearranged sequence reads was calculated.

Trough identification

To identify points of lower coverage inside peak areas, we converted the bam files of the immunoprecipitation libraries to BigWig files and subjected them to PeakSplitter59 with options of –valley 0.65 and –cutoff 30. The trough was identified as the minimal point ±10 nucleotides, between the summits of identified sub-peaks. Only troughs reoccurring in 3 biological repeats were considered.

Bidirectional conversion of genomic and transcriptomic coordinates

Genomic coordinates of all 5′ UTRs, CDS and 3′ UTRs of RefSeq transcripts were downloaded from the University of California, Santa Cruz genome browser (UCSC) table browser in BED format (hg19 and mm10 for human and mouse, respectively). Transcripts with IDs corresponding to more than one genomic location were discarded, as were transcripts of non-coding genes. Custom, in-house, awk and perl scripts were then used to convert transcriptomic coordinates to genomic coordinates and vice versa.

Definition of peak middle point

MACS2-generated genomic m1A peak coordinates were converted to transcriptomic coordinates. Neighbouring peaks (at a distance of up to 10 nucleotides) on the same transcript were merged using mergeBed from the BEDTools package60. This allowed the unification of peaks overlapping more than one exon that were divided by MACS2 due to the presence of introns in genomic coordinates. The middle of the resulting transcriptomic peak was defined as peak middle and referred to as m1A peak position in further analyses.

Metagene profiling

The relative position of each peak’s middle point was located in the corresponding transcript segment (5′ UTR, CDS or 3′ UTR). Each segment was assigned a value corresponding to its average length fraction out of the overall transcript length. The corresponding relative peak position was then deduced within each segment and plotted along a normalized metagene profile with respect to the TSS, AUG start codon and stop codon positions. The distribution of m1A peaks with respect to the canonical AUG start codon, TSS, first and nearest splice sites, and stop codon was also drawn in actual nucleotide distances.

Sliding window analysis of AUG start codon window

Structural features of a 300-nucleotide window centred on the AUG start codon were compared between methylated and non-methylated RefSeq genes. GC content was calculated as the average percent in a sliding window of 3 nucleotides. Local free energy (ΔG) was calculated in a 30-nucleotide consecutive jumping window (in steps of 10 nucleotides) using the RNAfold tool in the ViennaRNA package61 with default parameter settings.

Minimum free energy (MFE) calculation

The 5′ UTR sequence of the predominant isoform of each gene (the highest expressed isoform in the input sample) was used to calculate GC content, MFE, length-adjusted MFE (aMFE) and MFE density (MFEden) using the ViennaRNA package61 and as previously described62.

PARS-seq and PARS scores calculations

Parallel analysis of RNA structure (PARS) was performed as previously described63 on HepG2 polyA+ RNA. Reads were mapped to the human transcriptome (UCSC known canonical genes) using Bowtie252 with local alignment option (--local). Only uniquely aligned reads were used for analysis. Read start positions were counted for each base in the transcriptome, as the read start is +1 to the nuclease cleavage site and therefore informative regarding local RNA 2D structure64. Only reads with more than 5 consecutive bases matching the transcript at the start of the read were counted. Read counts were normalized to the respective library size. PARS-scores were then calculated based on the following equation: V1′ ij and S1′ ij are the normalized read-start counts at base (i) in transcript (j) for V1 or S1, respectively.

PARS scores at the region around the start-codon (±150 nucleotides from the AUG) were averaged and used for a metagene analysis.

Motif search

To search for a motif, 101-nucleotide sequences in the AUG start codon window, centred on m1A peaks in HeLa and mouse liver, were used. Compatible background sequences were used to prevent skewing of the results due to over-representation of known motifs in the region of the start codon, such as Kozak sequences and the AUG. The findMotifs program from the HOMER package65 was then run using the settings -rna -len 6,7,8,9,10 –noweight. The same peak areas were used with MEME66 for de novo motif enrichment analysis, with the settings -dna -minw 6 -maxw 10 -evt 0.01 -maxsize 1,000,000. A similar analysis was performed on m1A peaks falling in the CDS with background sequences randomly generated from non-methylated exons. 199 sequences of 20 nucleotides, surrounding the mutations identified above, were used to create a frequency plot of the nucleotides surrounding the mutation sites using WebLogo67.

Comparison of m6A and m1A profiles

m1A and m6A peaks1,42,68 were intersected using BEDTools60. Similarity between sets was also evaluated by comparing the metagene profiles of the two modifications.

TIS correlation

TISs data was downloaded from http://www.ncbi.nlm.nih.gov/sra?term=SRA160745 (ref. 21) and genomic locations were converted to transcriptomic locations (as described above). For each alternative TIS, the closest m1A peak was identified and the distance of both from the canonical AUG start codon was calculated. Correlation of alternative TISs with the number of m1A peaks per gene was carried out by dividing the genes containing alterative TISs to bins according to the number of peaks per gene.

Gene ontology (GO) enrichment

Methylated gene RefSeq IDs were uploaded to DAVID Bioinformatics Resources (http://david.abcc.ncifcrf.gov) and functional enrichment analysis was performed using all adequately expressed genes as background. Resulting enriched GO terms were restricted to fold enrichment ≥ 1.5, Bonferroni corrected P ≤ 0.005.

Human–mouse conservation

m1A mouse liver peaks were converted to the homologous coordinates in the human hg19 genome, using the LiftOver tool of the UCSC genome browser69. To assess the significance of these results we chose at random an equal number of 200 bp sized areas from within mouse genes that have a human orthologue. The areas were chosen from all gene segments in the same proportion as observed in the original data set. Randomly chosen areas were similarly converted and intersected with HepG2 m1A peaks. To calculate the statistical significance the randomization was repeated 1,000 times and Mann–Whitney U test was used.

ALKBH3 overexpression

Plasmids encoding wild-type and mutant (D193A) FLAG-tagged human ALKBH3 were transfected into 293 Freestyle cells using polyethylenimine (PEI; Polysciences, Inc.) at a ratio of 3 μl PEI per 1 μg plasmid DNA. After 24 h, cells were pelleted and washed once with cold PBS before being lysed in Cell Lysis Solution (5 Prime). mRNA was purified and digested for LC-MS/MS analysis as described in the ‘RNA purification’ and ‘m1A detection and quantitation’ sections above.

Translation efficiency (TE) and ribosome release score (RRS)

Sequencing data was downloaded: for HEK293, mouse liver and MEFs - SRA16074521, http://www.ncbi.nlm.nih.gov/sra?term=SRA160745; for mESCs - GSE3083919, http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE30839. Reads were aligned and FPKM values were calculated (see above). For TE calculation, the Ribo-Seq FPKM of the most highly expressed transcript of each gene was divided by the mRNA-Seq FPKM of the same transcript. Only transcripts with FPKM and Ribo-Seq FPKM above first quartile were considered. TE data for mESCs was downloaded as-is from the original paper19. TE and RRS for HeLa were calculated using our published data44. RRS values were calculated using mRNA-Seq and Ribo-Seq reads as described previously34,42.

Protein abundance

All protein abundance data analysed in this study was downloaded from the MOPED database (Multi-Omics Profiling Expression Data, https://www.proteinspire.org): HeLa, HepG2, HEK293 (Experiment ID: Geiger_MCP_2012)70, and mouse liver (Experiment ID: Baylor_mouse_liver_profiling), or from Supplementary Data (mESCs71 and MEFs72, corrected as described73). For all cell types, excluding mESCs, we used the normalized abundances as provided. For mESCs, raw intensities were divided by each leading protein molecular weight. Protein abundance data was then log 2 transformed and subjected to an analysis of variance (ANOVA) with mRNA expression (FPKM) percentile bins for start m1A and non-m1A genes. For MEFs, the published mRNA expression data was used along with the corresponding protein abundance. Genes with mRNA expression below the lowest expressed gene containing m1A or with FPKM < 0.2 were discarded.

Analysis of covariance (ANCOVA)