Determination of m6A and 5mC conductivity using single-molecule electrical detection
We investigated methylation site detection of synthesized RNA nucleotides using a single-molecule electrical resequencing method. The detection scheme was as follows (Fig. 1a) (Supporting information (SI): S1–S6, Figs. S1–S3). In the first step, the conductivity of each nucleotide in the synthesized RNA molecules was measured by sequentially reading across individual single nucleotides with nano-fluid integrated nano-gap devices (Fig. 1b), in which the nano-fluid strongly confined nucleotide translocation, resulting in guidance of the nucleotide molecules straight into the fluid region under direct current (DC) voltage across the gap electrode. The obtained conductance–time profiles represented the conductance sequence of each nucleotide in the synthesized oligonucleotide translocated through the gap electrodes. In the second step, the Phred base-calling method27,28 was used for each of the conductance–time profiles based on the conductivity of mono-nucleotides (SI: S5–S7) and determined the sequences of the oligonucleotides being translocated through the gap electrodes (Fig. S5). In this base-calling method, the conductance profiles of any detectable types of nucleotides, including methylated ones, are required. Therefore, in this study, we re-measured the characteristic conductance profiles of 5mC and m6A (SI: S3, Figs. S1–S3) and determined the conductance values of 5mC and m6A for 105 picosiemens (10−12 S = pS) and 92 pS, respectively (SI: S2, and Fig. S2). Together with the previous conductance results of RNA mono-nucleotide21, we found that the order of the conductance values is as follows: 5mC (105 pS) > rGMP (87 pS) > m6A (92 pS) > rAMP (67 pS) > rCMP (60 pS) > rUMP (36 pS) (Table 1) (Fig. S4). These values were used for base calling in the second step. In the third step, the sequences determined were mapped by assembly against the original sample sequences (SI: S6, and Figs. S5, S6). Based on the mapped sequences, the conductance profiles were obtained, and the methylation ratios in the sample nucleotides were evaluated, especially each of the cytidine and adenosine sites in the sample nucleotides.
Determination of m6A and 5mC modification ratios in miRNAs of colon cancer cells
We previously reported that some RNA base modifications are enhanced in cancer cells compared with normal cells; in particular, the detection of m6A in miRNA would be helpful in pancreatic cancer diagnosis6. In this study, we applied this method for estimation of the methylation rate in sample miRNAs extracted from cells of a typical colorectal cancer cell line (DLD-1). Of the DLD-1 miRNAs, miR-200c-5p (5′-CGUCUUACCCAGCAGUGUUUGG-3′) is strongly associated with cancer progression and metastasis29. To measure RNA base modification levels in miR-200c-5p (SI: S9, and Fig. S7a), mature miR-200c-5p was extracted from total RNA and captured by a complementary DNA (cDNA) probe attached to magnetic beads before measurement of the conductance profiles. Figure 2a shows the conductance plots of the captured miR-200c-5p sample, which is constructed from 2000 conductance signals (SI: S3, and Fig. S2). The average conductance levels of the captured sample miRNA were in agreement with those of the non-methylated synthesized miR-200c-5p oligonucleotide (Fig. 2b,c). This suggests that most of the captured RNA was comparable to miR-200c-5p.
Determination of miR-200c-5p base sequence. (a) Heat maps of RNA conductance plots for miR-200c-5p extracted from colorectal cancer cells (DLD-1). (b) Heat maps of synthesized miR-200c-5p, in which adenosine and cytosine are non-methylated. (c) Heat maps of RNA conductance plots of synthesized miR-200c-5p in which adenosines #7 and #13 are methylated. The x and y axes are the base position and conductance normalized to the conductance of guanine, respectively. (d) Enlarged conductance plots of the #7 position adenosine for non-methylated miR-200c-5p (left), captured miR-200c-5p (middle), and methylated miR-200c-5p (right). (e) Enlarged conductance plots of the #13 cytidine for non-methylated miR-200c-5p (left), captured miR-200c-5p (middle), and methylated miR-200c-5p (right).
Importantly, in the miR-200c-5p conductance profiles, larger conductive signals around 1.2 relative G (normalized to the electrical conductance of guanine) coexisted with the conductance of cytidine around 0.6 relative G at the cytidine sites (Fig. 2e). Since the conductance levels of the larger signals were comparable to that of 5mC, the larger signals were likely due to 5mC signals (Table 1). The ratio of the 5mC signal number to the C signal number was found to be 4.6% (4809/103,924) (SI: S9, Fig. S7b). This suggests that 4.6% of cytidine in the miRNA was methylated, which is comparable to the 3.0% methylation rate found previously in small RNAs in HCT116 cells by liquid chromatography (LC)–tandem mass spectrometry (MS/MS).
Similarly, we found larger conductive signals around 0.9 relative G at the adenosine sites (Fig. 2d), which were comparable to that of m6A (Table 1). The ratio of m6A signal number to A signal number was 2.9% (1921/67,154), suggesting that 2.9% of adenosine in the sample miRNA was methylated. The methylation ratio determined by our method is comparable to the 1.2% ratio found previously in small RNAs in HCT116 cells by LC–MS/MS1.
Dependence of 5mC modification rate on cytidine base position and consensus sequence
We evaluated differences in the methylation ratios (5mC/C) of the cytidine sites (#4, #8, #9, #10, and #13) in the sample miR-200c-5p nucleotide by signal assembly of the right length of read sequence before and after the cytidine base number to be evaluated. The 5mC methylation ratio was the ratio of the number of methylated nucleotides to the number of non-methylated nucleotides for a given site. In the captured miR-200c-5p, 5mC/C was 0.6% for #4 (10/1606), 1.3% for #8 (28/2084), 1.4% for #9 (43/3137), 3.7% for #10 (72/1952), and 18.5% for #13 (302/1633) (Fig. 3a). This suggests that cytidine #13 of miR-200c-5p is highly methylated. The methylated fragment containing the #13 cytidine in miR-200c-5p was also detected by MALDI–TOF MS/MS (SI: S10, Fig. S8). Therefore, this method enables us to evaluate the methylation status of each of the cytidine sites in a miRNA.
m6A and 5mC counts for single-molecule methylated miR-200c-5p. (a) 5mC modification rates in miR-200c-5p. In the second column, the sequences neighboring the methylated cytosines are shown. In the third column, the conductance histograms relative to those of guanine are shown. The black and blue lines represent the typical relative conductance values for C and 5mC, respectively (Table 1). (b) m6A modification rates in miR-200c-5p. In the second column, the sequences neighboring the methylated adenosines are shown. In the third column, the conductance histograms relative to those of adenosine are shown. The black and red lines represent the typical relative conductance values for A and m6A, respectively (Table 1).
The 5mC methylation of RNA is carried out by methyltransferases, such as NSUN2. The consensus sequences targeted by NSUN2 are CHG, CHH, and CG (H = A, C, or U)30. Of the cytidine methylation sites, #4 (CUU), #8 (CCC), and #9 (CCA) match the CHH consensus sequence, and #10 (CAG) and #13 (CAG) match the CHG consensus sequence. The methylation ratios for the CHH and CHG sites were 1.2% (81/6827) and 10.4% (374/3585), respectively. These results suggest that the methylation in the miR-200c-5p sample extracted from colorectal cancer cells was induced by the NSUN2 methyltransferase.
Dependency of m6A modification rate on adenosine base position and consensus sequence
We also evaluated the differences in methylation ratios (m6A/A) between the adenosine sites (#7, #11, and #14). In the captured miR-200c-5p, the methylation ratios (m6A/A) were found to be 10.9% (227/2074) for #7, 7.2% (118/1629) for #11, and 3.8% (34/905) for #14 (Fig. 3b). This suggests that #7 is the most highly methylated adenosine site. The fragmented methylation of #7 adenosine in miR-200c-5p was also detected by MALDI–TOF MS/MS (SI, S5, Fig. S2). Therefore, this method enables us to evaluate the methylation status of each adenosine site in a miRNA.
It has previously been reported that the m6A methylation of RNA is caused by methyltransferases such as METTL3 complex31, and the consensus sequence that it recognizes is RACH (R = A or G; H = A, U, or C). A three-base match (matching rate: 75%) to the RACH consensus sequence was found in the sequence neighboring the methylation site of #7 (UACC), and a two-base match (matching rate: 50%) was found in those of #11 (CAGC) and #14 (CAGU). The similarities to the RACH consensus sequence of the sequences neighboring the methylated adenosines suggest that miR-200c-5p methylation was induced by a METTL3 methyltransferase complex.
Correlation of m6A and 5mC
Finally, we evaluated the signals of A and C methylations coexisting on a single molecule. In this study, we focused on correlating the 5mC methylation of site #13 with m6A methylation (sites #7 and #11) because the 5mC methylation was highest at site #13 compared with the other 5mC modification positions (#4, #8, #9, and #10) in the miR-200c-5p sample, as shown in Fig. 3a and b, respectively. Figure 4a shows each methylation signal number and ratio for sites #7–13 of miR-200c-5p. Of the total 1936 signal numbers, the ratio of non-methylated signals to total numbers was 68.8% (1332/1936), and the m6A (#4 and #11) modified signal ratio was 10.8% (210/1936). The 5mC (#13) modified signal ratio was 15.6% (302/1936).
Proposed mechanism for formation of the miR-200c-5p epitranscriptome. (a) Mechanism of formation of the epitranscriptome in the whole miRNA. Epi-distribution for all 1729 signal numbers in the whole miRNA. The signal numbers for all sequence combinations containing non-methylated/methylated #7 (A), #11 (A), and #13 (C) are shown. (b) Epi-distribution for all 255 signal numbers in the m6A-antibody-captured miRNA, (c) Epi-distribution all 1,489 signal numbers of the miRNA in the supernatant after m6A-antibody capture (m6A non-captured miRNA). (d) Structures of NSUN2 and YTHDF proteins. The RsmB domain in NSUN2 is denoted by a green rectangle. The YTH domains are indicated by light blue rectangles. (e) Amino acid homology between NSUN2 and YTHDF YTH domains. (f) Amino acid comparison between the putative YTH domain of NSUN2 and the YTH domains of the YTHDF proteins. The amino acids shown in red match the YTH domain exactly, and the amino acids shown in blue have similar properties to the YTH domain. The amino acids in red squares recognize m6A. (g) Model of how miRNA 5mC modification is induced by m6A modification. (h) Epi-distribution for all 1729 signal numbers in the whole miRNA. (left two panels). The A-methylation rate (m6A/A) in 5mC-containing miRNA which binds NSUN2 was 20.1% (76/378) (first left panel). On the other hand, A-methylation rate (m6A/A) in miRNA which neither contain 5mC nor bind NSUN2 was 13.6% (210/1542) (second from the left panel). Epi-distribution for all 946 signal numbers in the NSUN2-captured miRNA, and for all 471 signal numbers in non-NSUN2 captured miRNA (right two panels). The A-methylation rate (m6A/A) in 5mC-containing miRNA which binds NSUN2 was 27.4% (356/1302) (first right panel). On the other hand, A-methylation rate (m6A/A) in miRNA which neither contain 5mC nor bind NSUN2 was 12.7% (60/471) (second from the right panel).
Importantly, both m6A and 5mC modified signal to total ratios were 4.8% (92/1936). This was the first time that simultaneous detection of m6A and 5mC in the same miRNA molecule had been achieved. These results suggest there is crosstalk between m6A and 5mC methylation. For instance, the ratio of 5mC methylation among m6A methylated signals was 30.4% (92/302), which was much larger than 18.5% of the ratio of 5mC methylated signals to the non-methylated signals (302/1634), and 20.3% of the 5mC modification to the total signal number (394/1936) (Fig. 4a). To confirm the methylation rates, we investigated the methylation rate of 5mC in RNA immunoprecipitated using an anti-m6A-antibody (targets m6A-containing total RNA) and non-captured RNA samples (RNA without m6A modifications). The C-methylation rate (5mC/C) in m6A-containing miRNA was 29.8% (76/255) (Fig. 4b) and 15.6% (233/1489) in the miRNA sample not containing m6A modifications (Fig. 4c); these values are comparable to 30.4% and 18.5%, respectively.
Together, these results suggest that the m6A modification of #7 and/or #11 promoted the 5mC modification of #13 in miR-200c-5p in colorectal cancer cells. Because the 5mC methylation rate is generally influenced by the activities of methylation/demethylation enzymes, our results imply the activities of 5mC methylation/demethylation enzymes are promoted/deactivated by m6A modifications in miR-200c-5p. As mentioned previously, cytosine is methylated in miR-200c-5p if it occurs in a motif recognized by NSUN2, suggesting that this m6A-dependent cytosine methylation may be caused by NSUN2. Therefore, we hypothesized that the NSUN2 protein has an amino acid sequence that recognizes m6A. We investigated amino acid homology between NSUN2 and YTHDF1, YTHDF2, and YTHDF3, which are known as m6A recognition proteins (Fig. 4d). We found that amino acids 535–578 of the NSUN2 protein have about 85% similarity with the YTH domain of the YTHDF protein family (Fig. 4e). Furthermore, NSUN2 also retains the amino acid (KS–WC) sequence32 that the YTHDF protein family requires for the recognition of m6A. These results suggest that NSUN2 has a YTH-like domain, which may promote cytosine methylation by recognizing m6A (Fig. 4f,g). To confirm the presence of NSUN2 in the vicinity of m6A, we performed the RNA immunoprecipitation using the NSUN2 antibody. The A-methylation rate (m6A/A) in 5mC-containing miRNA which binds NSUN2 was 27.4% (356/1302) (Fig. 4h: first right panel). On the other hand, A-methylation rate (m6A/A) in miRNA which neither contain 5mC nor bind NSUN2 was 12.7% (60/471) (Fig. 4h: second from the right panel). As expected, NSUN2-bound miRNAs tended to be high in m6A. Therefore, it was suggested that NSUN2 may recognize m6A and bind to RNA. Moreover, these results tend to be similar to those when measuring whole RNA (Fig. 4h: left two panels).
Overall, we measured the conductance profiles using a nano-fluid integrated nano-gap electrode device and successfully detected both A and C methylation sites simultaneously in sample RNA nucleotides extracted from cancer cell lines. The methylation positions were comparable to those determined by MALDI–TOF MS/MS. Furthermore, we evaluated the methylation ratios for each C and A site in the sequences and their relationship at the single-molecule level. These results suggest that the methylation ratio 5mC/C is facilitated by the presence of vicinal m6A methylation. This method is applicable for the comprehensive analysis of methylation site detection in the epitranscriptome, which will be useful for understanding these methylation events and their mechanisms, ushering in a new era in RNA biology.
