The dynamic N(1)-methyladenosine methylome in eukaryotic messenger RNA

Abstract

Gene expression can be regulated post-transcriptionally through dynamic and reversible RNA modifications. A recent noteworthy example is N(6)-methyladenosine (m(6)A), which affects messenger RNA (mRNA) localization, stability, translation and splicing. Here we report on a new mRNA modification, N(1)-methyladenosine (m(1)A), that occurs on thousands of different gene transcripts in eukaryotic cells, from yeast to mammals, at an estimated average transcript stoichiometry of 20% in humans. Employing newly developed sequencing approaches, we show that m(1)A is enriched around the start codon upstream of the first splice site: it preferentially decorates more structured regions around canonical and alternative translation initiation sites, is dynamic in response to physiological conditions, and correlates positively with protein production. These unique features are highly conserved in mouse and human cells, strongly indicating a functional role for m(1)A in promoting translation of methylated mRNA.

Extended Data Figure 1. Experimental conditions for detection, quantitation and sequencing of m¹A in mRNA

a, Schematic illustration of m¹A-to-m⁶A rearrangement (Dimroth rearrangement) under alkaline conditions at elevated temperatures. b, mRNA purification scheme before LC-MS/MS m¹A quantitation (left). Corresponding RNA electrophoresis profiles obtained by Agilent 2100 Bioanalyzer (middle and right). c, Monitoring m¹A-to-m⁶A rearrangement levels during sample preparation for LC-MS/MS. Synthetic 5-nucleotide long RNA oligonucleotides containing m¹A (upper panel) or m⁶A (lower panel) were digested to mononucleotides, dephosphorylated (see methods) and analysed by HPLC-UV. Only minimal (<10%) m¹A-to-m⁶A rearrangement was observed (arrow). d, Standard curves for m¹A and deuterium-labelled m¹A (d₃-m1A) demonstrate very similar detection sensitivity in LC-MS/MS. Mean values ± s.e.m. are shown, n = 3. e, The change in m¹A/A and m⁶A/A molar ratios (%) during the purification scheme outlined in b. Mean values ± s.e.m. are shown, n = 3. f, LC-MS/MS of mRNA isolated from HepG2 cells labelled with deuterated methionine (d₃-Met) for 24 h detects d₃-m1A, suggesting S-adenosyl-methionine (SAM) is the methyl donor. Mean values ± s.e.m. are shown, n = 3. g, Dot blots demonstrating high anti-m¹A antibody specificity. Increasing amounts (indicated above the top blot) of synthetic RNA oligonucleotides containing m¹A, m⁶A or unmodified A residues were spotted onto a membrane and probed with either anti-m¹A or anti-m⁶A antibodies. Anti-m¹A antibody detects m¹A and does not exhibit cross-reactivity with m⁶A or A (upper blot); anti-m⁶A antibody demonstrates low, yet detectable, cross-reactivity with both m¹A and A (lower blot). For blot source data, see Supplementary Fig. 1. h, Competitive dot blots were performed on separate membranes spotted with 75 pmol of synthetic m¹A-containing RNA oligonucleotide. Whereas increasing concentrations of free m¹A mononucleoside progressively attenuate anti-m¹A binding, increasing concentrations of free m⁶A mononucleoside do not. For blot source data, see Supplementary Fig. 1. i, Quantitative LC-MS/MS demonstrates m¹A enrichment—and m⁶A depletion—following immunoprecipitation (IP) with anti-m¹A antibody compared to total RNA input. Mean values ± s.e.m. are shown, n = 3. j, Monitoring m¹A-to-m⁶A rearrangement levels under different RNA fragmentation conditions for use in m¹A-seq. Pure m¹A mononucleoside in 1×fragmentation buffer (see Methods) was subjected to the conditions specified to the right of the chromatograms and directly analysed by injection to HPLC-UV for rearrangement to m⁶A. k, Comparison of competitive m¹A elution and Proteinase K elution of immunoprecipitated m¹A-containing RNA fragments from anti-m¹A-coupled magnetic beads shows that the two elution modes are equivalent. l, Identification of the known m¹A site in position 1322 of human 28S rRNA validates the accuracy of m¹A-seq and the use of peak middle points as a close approximation for m¹A sites. Partial m¹A-to-m⁶A rearrangement increases the coverage around this site. m, Conditions for induced m¹A-to-m⁶A rearrangement of RNA oligonucleotides that maintain RNA integrity for use in m¹A-seq. A synthetic 5-nucleotide long RNA oligonucleotide of the sequence 5′-AC(m¹A)UG-3′ was subjected to various base/heating conditions (indicated to the right of the chromatograms) and directly analysed by injection to HPLC-UV for rearrangement to 5′-AC(m⁶A)UG-3′. Incubation at pH 10.4, 60 °C for 1 h results in rearrangement to m⁶A in 40% of oligonucleotides; longer incubation times result in increased rates of rearrangement. Chromatograms of untreated RNA oligonucleotides appear above and mark the expected retention times.

Extended Data Figure 2. Characterization of human m¹A-methylated genes

a, Gene ontology (GO) analysis of methylated HeLa genes relative to all adequately expressed genes (above the 1st quartile) reveals enrichment of biological processes related to translation and RNA metabolism. Fold-enrichment and P values are indicated for each category. b, GO analysis of molecular functions reveals enrichment of structural constituents of the ribosome. Scheme based on an illustration obtained from DAVID bioinformatics website of the KEGG human ribosome pathway. Red stars indicate methylated genes in the pathway. Colouring of the boxes and ribosome constituents is according to KEGG pathway maps showing interacting proteins and hyperlinks to gene entries that can be reached through http://www.genome.jp/kegg-bin/show_pathway?hsa03010. c, The average fraction of methylated transcripts (stoichiometry) increases with gene expression level. r and P values are indicated, Pearson correlation. d, Pie chart presenting the fraction of m¹A peaks in each of three non-overlapping transcript segments (5′ UTR, CDS and 3′ UTR) in HEK293 cells. e–g, The fraction of methylated genes increases with gene expression levels in HepG2 (e), HEK293 (f) and common human peaks (see Methods) (g).

Extended Data Figure 3. Metagene analyses of human m¹A peaks

a, Metagene profiles demonstrating sequence coverage along a normalized gene transcript. Sequence reads of m¹A immunoprecipitation and input in HeLa cells are indicated in blue and orange, respectively. b–d, Metagene profiles of m¹A peak distribution in a non-normalized window centred on the AUG start codon (b), extending downstream from the transcription start site (TSS) (c) and centred on the stop codon (d), in the indicated human cell types. e, Metagene profiles comparing the distribution of m¹A peaks (blue) to that of negative peaks (black) along a normalized transcript composed of three rescaled non-overlapping segments illustrated below, in HeLa cells. f, Table demonstrating m¹A peak enrichment in growing windows centred on the AUG start codon in HeLa cells. Enrichment is calculated as number of peaks in the window divided by window size (nucleotides). g, Table summarizing the overlap between m⁶A and m¹A peaks in HepG2, HEK293 and WT mESCs. m⁶A peaks are sourced from Dominissini et al., Linder et al. and Geula et al., respectively. m¹A peaks are from the current study. h, i, Metagene profiles of m¹A peak distribution in a non-normalized window centred on the AUG codon (h) and extending downstream from the TSS (i). Peaks are sorted by the exon containing the AUG codon and the length of the first exon, respectively. j, Metagene profiles of m¹A peak distribution in a non-normalized window centered on the nearest splice site. Peaks are sorted by the exon containing the AUG codon in that gene. k, m¹A-induced reverse transcription (RT) arrests produce typical m¹A peaks characterized by a central region of reduced coverage with a local minimum (m¹A trough) Examples are shown. l, m¹A-to-m⁶A rearrangement results in a reduced number of identified m¹A troughs. Higher rates of rearrangement further reduce the number of identified m¹A troughs (right panel). Example is shown (left panel). m, Metagene profile of m¹A trough distribution along a normalized transcript composed of three rescaled non-overlapping segments illustrated below, in HepG2 cells. n, Metagene profile of m¹A trough distribution in a non-normalized window centred on the AUG start codon in HepG2 cells. o, Metagene profile of m¹A trough distribution along a normalized transcript in HepG2 cells. p, Metagene profile of m¹A trough distribution in a non-normalized window centred on the nearest splice site in HepG2 cells. q, Pie chart presenting the fraction of m¹A troughs in each of three non-overlapping transcript segments (5′ UTR, CDS and 3′UTR) in HepG2 cells. r, MEME motifs identified in 100-nucleotide windows centred on m¹A troughs that lie within the AUG start codon window (±150 nucleotides) in HepG2 cells.

Extended Data Figure 4. Correlation between m¹A sites and TISs in human

a, b, The mean number of alternative TISs in methylated (m¹A) versus all other (unmethylated) genes in HeLa (a) and HEK293 (b) cells. Mean values ± s.e.m. are shown. P values are indicated, Mann–Whitney U test. c, The mean number of alternative TISs per gene as a function of the number of m¹A peaks per gene in HeLa and HEK293 cells. Mean values ± s.e.m. are shown; r and P values are indicated, Pearson correlation; regression line is drawn. d, e, The percentage of genes with upstream (=5′ UTR) or downstream (=CDS) m¹A sites out of all genes that have either downstream or upstream TISs, compared to the expected percentage in HeLa (d) and HEK293 (e) cells. P values are indicated, χ² test. f–h, Scatter plots showing the correlation between the locations (log₂) of alternative TISs and m1A peaks with respect to the canonical AUG start codon (0) in HeLa (f), HEK293 (g) and HepG2 (h) cells: left, upstream TISs (uTIS) and 5′ UTRs m¹A peaks; right, downstream TISs (dTIS) and CDS m¹A peaks. r and P values are indicated, t-test.

Extended Data Figure 5. Structure and sequence features of m¹A sites in human mRNA

a, m¹A peaks have a significantly higher GC content compared to negative peaks in all three transcript segments: 5′ UTR, CDS and 3′ UTR. Box limits represent 25th percentile, median and 75th percentile, whiskers represent 2.5 and 97.5 percentiles, and dots indicate outliers. P = 1.5 ×10 ⁻²⁷⁸, P = 8.2 ×10 ⁻²⁵⁹ and P = 3.3 ×10 ⁻²⁷¹, respectively, t-test. b, Motifs identified in 400-nucleotide windows centred on the canonical AUG start codon in genes with m¹A peaks in this window (upper table), or around m¹A peaks located in the CDS, outside the AUG start codon window (lower table). c, Examples of adenosines around m¹A peak middles with increased mismatch rates. Fold-enrichment values are the ratios of mismatch rates in untreated relative to rearranged samples. The top ten highest fold enrichment samples are shown. d, e, GC content (d) and minimum free energy density (MFEden) (e) of 5′ UTRs of methylated (m¹A), unmethylated (Non-m¹A) and all genes. Box limits represent 25th percentile, median and 75th percentile, whiskers represent 2.5 and 97.5 percentiles. P values are indicted, t-test. f, Mean GC content (upper panel), PARS score (middle panel) and free energy (ΔG, lower panel) in a 300-nucleotide window centred on the start codon of commonly methylated genes relative to non-methylated genes (see Methods). Error bars represent s.e.m. g, The PARS scores of methylated compared to all other genes in HepG2 cells in a 150-nucleotide window extending downstream from the TSS, calculated in 30-nucleotide sliding windows. Each plot represents data from an independent PARS experiment. Error bars represent s.d.

Extended Data Figure 6. Features of the mouse m¹A methylome (part 1)

a, Detection of an m¹A site at position 1135 of mouse 28S rRNA, in mouse liver m¹A immunoprecipitation. A drop in sequence read coverage can be seen at the methylated position. b, Fold-enrichment (immunoprecipitation over input reads) identifies an m¹A peak. c, High mismatch rate at the identified m¹A 1135 in mouse 28S rRNA. d, Pie charts presenting the fraction of m¹A peaks in each of three non-overlapping transcript segments (5′ UTR, CDS and 3′ UTR) in the indicated mouse cell types. e, Metagene profiles of m¹A peak distribution in a non-normalized window centred on the AUG start codon in the indicated mouse cell types. f, Table showing m¹A peak enrichment in growing windows centred on the AUG start codon in mouse liver. Enrichment is calculated as the number of peaks in the window divided by window size (nucleotides). g, h, Metagene profiles of m¹A peak distribution in a non-normalized window extending downstream from the TSS (g) and centred on the stop codon (h) in the indicated mouse cell types. i, The mean number of alternative TISs in methylated (m¹A) versus all other (unmethylated) genes in MEF cells. Mean values ± s.e.m. are shown; P value is indicated, Mann–Whitney U test. j, The percentage of genes with upstream (=5′ UTR) and downstream (=CDS) m¹A sites out of all genes that have either downstream or upstream alternative TISs, compared to the expected percentage in MEF cells. P value is indicated, χ² test. k, l, Scatter plots showing the correlation between the locations (log₂) of upstream TISs (uTIS) and 5′ UTR m¹A peaks relative to the canonical AUG start codon (0) in mouse liver (k) and MEF (l) cells. r and P values are indicated, t-test. m, Metagene profiles of mouse m¹A peak distribution in a non-normalized window centred on the AUG codon. Peaks are sorted by the exon containing the AUG codon in that gene.

Extended Data Figure 7. Features of the mouse m¹A methylome (part 2)

a, Mouse m¹A peaks have a significantly higher GC content compared to negative peaks. Box limits represent 25th percentile, median and 75th percentile, whiskers represent 2.5 and 97.5 percentiles, and dots indicate outliers. P =4.4 ×10 ⁻¹⁷⁵, t-test. b, Motifs identified in 400-nucleotide windows centred on the canonical AUG start codon in genes with m¹A peaks in this window in mouse liver. c, d, GC content (c) and MFEden (d) of 5′ UTRs of methylated (m¹A), unmethylated (Non-m¹A) and all genes. Box limits represent 25th percentile, median and 75th percentile, whiskers represent 2.5 and 97.5 percentiles. P values are indicted, t-test. e, f, A sliding window profile of mean GC content (e) and mean ΔG (f) in a 300-nucleotide window centred on the canonical AUG start codon in methylated (m¹A) genes compared to all other genes in mouse liver. P values are indicated, Kolmogorov–Smirnov test and t-test. g, Representative plots of human-mouse orthologous genes with conserved m¹A peaks. Plot format as in Fig. 2a.

Extended Data Figure 8. The yeast m¹A methylome

a, m¹A-seq identifies the known m¹A sites (645 and 2142) in S. cerevisiae 25S rRNA. A drop in sequence read coverage (indicated by purple dots) occurs around the methylated positions (indicated by dashed lines). b, Representative plots of methylated transcripts in S. cerevisiae. Plot format as in Fig. 2a. c, The percentage of methylated genes that carry 1, 2 or 3 m¹A peaks per gene in S. cerevisiae. Out of 843 m¹A peaks (FC ≥4, FDR ≤ 5%) in 778 genes, most (88.6%) are methylated only once. Unlike in mammals, m¹A is distributed across the coding transcriptome without an apparent preferred location. d, m¹A-seq identified the known m¹A sites (670 and 2230) in S. pombe 25S rRNA. A drop in sequence read coverage (indicated by purple dots) occurs around the methylated positions (indicated by dashed lines). e, Representative plot of a methylated transcript in S. pombe. Plot format as in Fig. 2a. f, Pie charts presenting the fraction of m¹A peaks in each of three non-overlapping transcript segments (5′ UTR, CDS and 3′ UTR) in S. pombe Sp1 strain under the indicated conditions. Under vegetative growth, we identified 706 m¹A peaks (FC ≥4, FDR ≤ 5%) in 619 gene transcripts, most of which (90.4%) distributed along the CDS. Four hours after transfer to a nitrogen-source deficient ‘sporulation’ medium, 157 out of the vegetative state m¹A peaks were no longer detected, and 297 new peaks appeared. Importantly, transcripts that harbour differential peaks were adequately expressed (above the 1st quartile) in both conditions. g, The percentage of methylated genes that carry 1, 2, or 3 +m ¹ A peaks per gene in S. pombe Sp1 strain under the indicated conditions. h, Venn diagram representing differential and shared m¹A peaks in S. pombe Sp1 strain under the indicated conditions. i, Representative plots of a differentially methylated transcript in S. pombe Sp1 strain under the indicated conditions. Yellow box, conserved peak; green box, differential peak.

Extended Data Figure 9. m¹A in mRNA is a dynamic modification

a, b, LC-MS/MS quantification of m¹A in mRNA of untreated and amino acid (AA)-starved (a) or serum-starved (b) HepG2. Mean values ± s.e.m. are shown, n = 3, *P ≤ 0.05, ***P ≤ 0.001, unpaired t-test. c, d, Representative plots of differentially methylated transcripts in untreated and glucose-starved (c) or heat shock-treated (d) HepG2 cells. Plot formats as in Fig. 2a. e, LC-MS/MS quantification of m¹A in mRNA of 293F cells overexpressing WT FLAG-ALKBH3 or an inactive mutant (D193A), presented as percentage of unmodified A. Mean values ±s.e.m. are shown, n = 3, *P ≤ 0.05, NS, P > 0.05, one-way ANOVA with Dunnet’s multiple comparison test.

Extended Data Figure 10. m¹A around the start codon correlates with higher translation efficiency (TE)

a, Cumulative distribution of log₂(TE) in genes methylated in a 300-nucleotide window centred on the start codon compared to all other genes, in the indicated human and mouse cell types. P values (t-test) and fold-changes (FC) of median TE values (Start m¹A genes/All the rest) are indicated. b, Genes methylated in a 300-nucleotide window centred on the AUG start codon have a higher ribosome release score (RRS = TE[CDS]/TE[3 ′ UTR]) compared to all other genes in the indicated cell types. RRSs, which are ‘normalized’ to ribosomal drop-off in the 3′ UTR, are in line with TE scores. P values (Mann–Whitney U test) and fold-changes (FC) of median RRS values (Start m¹A genes/All the rest) are indicated. c, Genes methylated in different start codon window sizes have higher TE compared to all other genes, in HeLa cells. When considering all m¹A genes, including those methylated outside the start codon window, the effect is reduced. P values (t-test) and fold-changes (FC) of median TE values (Start m¹A genes/All the rest) are indicated. Box limits represent 25th percentile, median and 75th percentile; whiskers extend from the limit to the highest and lowest value within 1.5 IQR (interquartile range).

Figure 1. Development of m¹A-seq to map a newly identified constituent of mammalian mRNA

a, Chemical structures of m¹A and m¹A is in blue. b, LC-MS/MS quantitation of m¹A, m⁶A and Ψ in human and mouse mRNA isolated from the indicated cell types. The level of each modified nucleoside is indicated as a percentage of the unmodified one. Mean values ± s.e.m. are shown, n = 3. MEFs, mouse embryonic fibroblasts; mESC, mouse embryonic stem cells; M3 KO, Mettl3 knockout. c, Schematic outline of m¹A-seq. RNA is fragmented and subjected to immunoprecipitation using anti-m¹A antibody. Eluted RNA fragments are converted to cDNA and sequenced, or treated to induce partial m¹A-to-m⁶A rearrangement before cDNA synthesis. m¹A causes both reverse transcription stops and read-throughs accompanied by mismatches, to produce typical peaks with a central trough and an adenosine with increased mismatch rate (left). Partial rearrangement of m¹A to m⁶A attenuates the effect (right). m¹A is in red, m⁶A is in grey; mismatch rate is illustrated as the ratio between A (red) and T (black). d, Detection of the known m¹A 1322 site (yellow) in human 28S rRNA validates m¹A-seq (black curve). High fold-change in mismatch rate (before and after m¹A-to-m⁶A rearrangement in immunoprecipitation, blue lines) independently identifies m¹A 1322 with single-nucleotide resolution. e, Mismatch rates in m¹A 1322 increase after immunoprecipitation and decrease upon rearrangement.

Figure 2. m¹A is associated with translation initiation sites (TISs) in the human transcriptome

a, Representative plot of a methylated transcript. Normalized sequence coverage of immunoprecipitation (m¹A IP, red) and input (blue) are indicated above the gene architecture in UCSC format. Thin black boxes represent the 5′ and 3′ untranslated regions (UTRs), thick black boxes represent the coding sequences (CDS) and thin lines represent introns. b, The percentage of methylated genes that carry 1, 2, 3 or 4+ peaks per gene in the indicated human cell types. Most methylated genes carry only one m¹A peak. c, The percentage of methylated genes in HeLa cells increases with expression level. d, Pie charts presenting the fraction of m¹A peaks in each of three non-overlapping transcript segments (5′ UTR, CDS and 3′ UTR) in the indicated human cell types. e, Stoichiometry (fraction) of methylated transcripts in genes that carry one m¹A peak in HepG2 mRNA. Plotted is the number of genes sorted by percentage of methylated transcripts. Dashed lines define the stoichiometry interquartile range (17–23%). f, Metagene profiles of m¹A peak distribution along a normalized transcript composed of three rescaled non-overlapping segments illustrated below, in the indicated human cell types. m¹A peaks cluster around the AUG start codon. For comparison, the distribution of m⁶A peaks in HepG2 is superimposed. g, Metagene profiles of m¹A peak distribution in a non-normalized window centred on the first splice site. Peaks are sorted by the exon containing the AUG codon in that gene.

Figure 3. m¹A occurs in GC-rich sequence contexts and in genes with structured 5′ UTRs

a, Sequence frequency logo for a set of 192 adenosines in peak areas that have a higher mismatch rate in immunoprecipitation relative to input (FC ≥ 6) in HepG2 demonstrates the GC-rich context of m¹A. b, Length-adjusted minimum free energy (aMFE) for 5′ UTRs of methylated, unmethylated and all genes. Box limits represent 25th percentile, median and 75th percentile, whiskers represent 2.5 and 97.5 percentiles. P values are indicated, t-test. Analysis was based on human common peaks (see Methods). c–e, A sliding window of mean GC content (c), ΔG (d) and PARS scores (e) in a ±150-nucleotide window centred on the AUG start codon of methylated genes compared to all the rest, in HepG2 cells. P values are indicated; Kolmogorov–Smirnov test, t-test and Kolmogorov–Smirnov test, respectively.

Figure 4. m¹A methylome conservation between human and mouse

a, The percentage of methylated genes that carry 1, 2, 3 or 4+ peaks per gene in the indicated mouse cell types. Most genes carry only one m¹A peak. b, The percentage of methylated genes in mouse liver cells increases with expression level. c, Pie chart presenting the fraction of m¹A peaks in each of three non-overlapping transcript segments (5′ UTR, CDS and 3′ UTR) in mouse liver. d, Metagene profiles of m¹A peak distribution along a normalized transcript composed of three rescaled non-overlapping segments illustrated below, in the indicated mouse cell types. m¹A peaks cluster around the AUG start codon. e, Metagene profiles of mouse m¹A peak distribution in a non-normalized window centred on the nearest splice site. Peaks are sorted by the exon containing the AUG codon in that gene. f, Length-adjusted minimum free energy (aMFE) for 5′ UTRs of methylated, unmethylated and all genes in mouse liver. Box limits and whiskers are as indicated in Fig. 3b. P values are indicated, t-test. g, Human–mouse m¹A conservation expressed as per cent of orthologous positions with shared m¹A peaks according to their location in the transcript.

Figure 5. m¹A in mRNA is a dynamic modification that responds to changing physiological and stress conditions, and varies between tissues

a, LC-MS/MS quantification of m¹A (left, grey) and m⁶A (right, black) in mRNA of untreated and glucose-starved (upper panels) or heat shock-treated (lower panels) HepG2 cells, presented as percentage of unmodified A. Mean values ± s.e.m. are shown, n = 3, **P ≤0.01, not significant (NS) P >0.05, unpaired t-test. b, Venn diagrams of differential and shared m¹A peaks in untreated and glucose-starved (left upper panel) or heat shock-treated (left lower panel) HepG2 cells. The overall change in the number of identified m¹A peaks is presented as per cent of untreated (right upper and lower panels). c, Representative plots of differentially methylated transcripts in untreated and glucose starved (upper panel) or heat-shock-treated (lower panel) HepG2 cells. Plot format as in Fig. 2a. Yellow boxes frame differential peaks. d, LC-MS/MS quantification of m¹A in the indicated wt/wt and ob/ob mouse tissues, presented as percentage of unmodified A. Mean values ± s.e.m. are shown, n = 3, *P ≤ 0.05, NS, P > 0.05, one-way ANOVA.

Figure 6. m¹A around the start codon correlates with higher protein levels

a–f, Genes methylated in a 300-nucleotide window centred on the start codon have higher protein levels than all other genes with similar RNA expression levels in HepG2 (a), HEK293 (b), HeLa (c), mESC (d), MEFs (e) and mouse liver (f). ANOVA m¹A P values against log₂ protein levels and the average fold-change (Avg. FC) of median protein levels across all gene expression bins (Start m¹A/Non-m¹A) are indicated. FPKM, fragments per kilobase of transcript per million mapped reads.