ABSTRACT
N6-methyladenosine (m6A) is a modified nucleotide found in mRNA, ribosome RNA (rRNA) and small nuclear RNA (snRNA). m6A in mRNA has important roles in regulating mRNA stability, splicing, and other processes. Numerous studies have described m6A as a dynamic modification using mass spectrometry-based quantification of m6A in mRNA samples prepared from different cellular conditions. However, these results have been questioned based on the finding that the mRNA purification protocols often result in varying levels of rRNA contamination. Additionally, mRNA purification protocols disproportionately enrich for the 3’ ends of mRNA, a region that is enriched in m6A. To address these problems, we developed the Phospho-tag m6A assay, a highly efficient method for quantifying m6A specifically from mRNA. In this assay, a series of selective RNase digestion steps is performed, which results in m6A from rRNA and snRNA being liberated as m6A monophosphate, while m6A from mRNA is mostly liberated as m6A nucleoside. m6A levels are normalized to transcript levels, using m7G monophosphate liberated by yDcpS decapping enzyme as a surrogate for mRNA levels. Notably, this approach uses total cellular RNA, rather than purified mRNA, which simplifies the steps for m6A detection and overcomes the 3’-end biases associated with mRNA purification. Overall, the Phospho-tag m6A provides a simple and efficient method for quantification of mRNA-derived m6A from total RNA samples.
Introduction
A major concept is that N6-methyladenosine (m6A), the most abundant internal nucleotide modification in mRNA, is “dynamic,” i.e., its levels can change in mRNA in different cell types or conditions. For example, changes in the levels of METTL3/METTL14, the biosynthetic heterodimeric methyltransferase that synthesizes m6A in mRNA1,2, are thought to lead to alterations in m6A levels in mRNA in specific disease states or cellular contexts3. These changes in m6A levels may affect the stability or processing of specific m6A-modified transcripts, thus affecting diverse cellular pathways and processes4.
The concept that m6A is dynamic comes from various types of m6A quantification assays, including dot blots and mass spectrometry, the latter of which is considered more quantitative. However, questions have been raised about the accuracy of these methods5,6,7. The major problem is that these methods require pure mRNA8. mRNA is typically <2%of total cellular RNA, while other m6A-containing RNAs such as snRNA, and especially rRNA, are ∼70% of total cellular RNA9. As a result, even highly efficient mRNA enrichment methods result in mRNA preparations that contain variable amounts of residual rRNA or snRNA10. For example, several studies demonstrated large amounts of residual rRNA levels even after two rounds of oligo-dT-based purification of mRNA6. rRNA-removal kits similarly leave considerable amounts of rRNA in samples11. Overall, these studies demonstrate that ostensibly pure mRNA fractions usually contain variable amounts of rRNA. Thus, putative differences in m6A in different samples could simply reflect differences in rRNA contamination.
A second major problem with using purified mRNA for m6A quantification is that mRNA is rarely purified as a full-length transcript. This is because RNA is highly susceptible to degradation by nucleases during purification. As a result, when mRNA is purified using oligo-dT resin, the resulting RNA fragments are biased for the 3’ end of the mRNA that contains the poly(A) tail12. This 3’ bias has been widely documented and seen to varying degrees in RNA-Seq studies13. Since m6A is enriched in mRNA 3’UTRs14,15, sample-to-sample variability in 3’ bias can result in different levels of m6A even if there are no underlying differences in methylation between samples. Overall, these highly prevalent artifacts of mRNA purification raise questions about the accuracy of previous m6A measurements which have lead to the conclusion that m6A is dynamic in the transcriptome.
To develop a method to reliably quantify m6A without the biases caused by mRNA purification, we developed the “phospho-tag” m6A assay. The phospho-tag m6A assay uses a simple and highly efficient sequential digestion protocol and relies on total cellular RNA rather than purified mRNA. The m6A that is liberated from the cellular RNA contains either a 5’-hydroxyl or a 5’-phosphate. m6A with a 5’-hydroxyl exclusively derives from mRNA, while m6A with a 5’-phosphate (i.e., m6AMP) derives from the alternative cellular m6A sources, i.e., rRNA and snRNA. Because these two forms of m6A have different masses, they can be readily resolved and quantified by mass spectrometry. In order to normalize mRNA-derived m6A levels, we developed a second approach. This method quantifies the total number of mRNA transcripts in a cell using a similar a phosphate-tagging method: m7G derived from mRNA caps contains a 5’-phosphate (i.e., m7GMP), while m7G derived from tRNA/rRNA contains a 5’ hydroxyl. The phospho-tag m6A assay is a “one-pot” reaction that simultaneously generates both m6A and m7GMP for quantification of m6A normalized to total mRNA levels. Together, the phospho-tag m6A assay provides a simple and rapid method for highly accurate quantification of mRNA-derived m6A from essentially any biological sample.
RESULTS
rRNA is highly prevalent in MS samples despite mRNA enrichment protocols
A potential problem with mass spectrometry-based m6A measurements is the possibility that rRNA is present as a contaminant in the poly(A) RNA fraction. mRNA purification using either oligo dT or rRNA depletion oligonucleotides has been shown to produce mRNA that can possibly have considerable levels of rRNA contamination10,11. If there is rRNA in these mRNA samples, this would be problematic since rRNA is known to contain m6A16,17 and could therefore account for the m6A detected by mass spectrometry. Notably rRNA contamination would lower the calculated m6A prevalence in a sample since m6A levels in rRNA is lower than m6A levels in mRNA. The m6A prevalence is 0.23% of adenosines in 18S rRNA and 0.23% in 28S rRNA (1 m6A in the 1869 nt 18S rRNA and 1 m6A in the 5070 nt 28S rRNA). In contrast, m6A prevalence in mRNA is approximately twice as high—the level of m6A in mRNA is reported to be 0.4% of adenosines14,15. Thus, variation in the degree of rRNA contamination between samples could give the erroneous conclusion that m6A in mRNA is actually changing between experimental conditions.
We first wanted to determine if rRNA contamination is a problem when analyzing poly(A) mRNA samples. To test this, we used published RNA-Seq and m6A-Seq datasets in which the purified mRNA was used in mass spectrometry analysis for m6A and subsequent RNA-Seq and m6A-Seq analysis. In each of the examined datasets, the authors used either two rounds of oligo(dT) purification, or they coupled oligo(dT) purification to rRNA depletion methods18,19.
To determine if rRNA was present in these samples, we downloaded the raw sequencing reads from the deposited RNA-Seq and m6A-Seq datasets and aligned them to the human and mouse rDNA sequences to quantify rRNA-mapping reads, which are normally removed in most alignment protocols20. Here we found that rRNA reads accounted for a substantial fraction of total RNA-Seq and m6A-Seq reads, ranging from 10% to 60% in these datasets from different studies (Fig. 1A). The presence of rRNA in these samples highlights the difficulty in removing rRNA despite the high attention given to mRNA purification.
We next examined RNA-Seq and m6A-Seq datasets from experiments that reported changes in m6A levels in mRNA, such as a recent study which found elevated m6A in acute myeloid leukemia18 and decreased m6A during stress response in major depressive disorder (MDD)19. We again examined the RNA-Seq datasets that used poly(A) RNA prepared in the same way as the poly(A) RNA used for mass spectrometry. Here, in these datasets, which were reported to contain higher m6A levels18,19, exhibited notably lower rRNA levels in input RNA-Seq dataset (Fig. 1A).
Notably, we were unable to find studies where the authors addressed the possibility of rRNA contamination by measuring the rRNA levels. Thus, rRNA contamination could account for the observed “dynamic” changes in m6A levels in mRNA. Overall, these data highlight the difficulty in removing rRNA from mRNA samples, and the need to develop new methods that overcome this unavoidable contamination.
Phospho-tag m6A assay: A method for selective detection of m6A mediated by METTL3/14 in total cellular RNA
Because obtaining highly pure mRNA samples is probably unrealistic for most laboratories, we wanted to develop an m6A quantification protocol that does not require mRNA purification. Thus this protocol must selectively measure mRNA-derived m6A in total RNA samples despite the vastly higher levels of rRNA and snRNA, which also contain m6A.
To measure m6A in mRNA, we took advantage of the unique sequence context of m6A in mRNA and other RNA polymerase II-derived transcripts. In mRNA, m6A is usually, but not always, preceded by a G21,22. This sequence preference reflects the methylation specificity of METTL3/METTL14, the heterodimeric enzyme that synthesizes m6A23,24,25. m6A mapping studies, along with earlier biochemical analysis, found that ∼75% of m6A is in a G-m6A-C sequence context, while 20-25% of m6A can be found in a A-m6A-C sequence context21,22,26. Currently, there are no known pathways that would selectively methylate m6A in a G-m6A-C context versus an A-m6A-C context. For this reason, m6A in a G-m6A-C context could serve as a proxy for the total level of m6A in the mRNA transcriptome.
Importantly, m6A in possible contaminating RNA is found in different sequence context. For rRNA, m6A is found exclusively in a A-m6A context27. In the U6 snRNA, m6A is found in a C-m6A context28. Therefore, any m6A in a G-m6A context reflects m6A from mRNA, not a contaminant.
In contrast, m6A in 18S and 28S rRNA is synthesized by METTL5 and ZCCHC4, respectively29,30, while m6A in U6 snRNA is catalyzed by METTL1631. Since only m6A in mRNA can be preceded by a G, the G-m6A sequence context can be used as a selective mark for m6A levels in mRNA.
To develop an assay to selectively quantify mRNA-derived m6A, we therefore took advantage of RNase T1, which selectively cleaves after G. Unlike other ribonucleases, RNase T1 leaves a 5’ hydroxyl after cleavage (Fig. 1B). Therefore, all m6A residues in a G-m6A-C sequence context are cleaved to HO-m6A-C-N (where N indicates one or more residues). Next, nuclease S1 is added, which cleaves all remaining phosphodiester bonds, including A-m6A in rRNA and C-m6A in snRNA. Nuclease S1 cleavage leaves a 5’ phosphate. As a result of nuclease S1, m6A residues in rRNA and snRNA contain a 5’ phosphate, i.e., m6AMP (Fig. 1B). However, cleavage of the RNase T1 digestion product OH-m6A-C-N by nuclease S1 liberates the non-phosphorylated m6A.
Overall, this “phospho-tag” assay is expected to generate two forms of m6A, i.e., m6A with a 5’-hydroxyl or a 5’ phosphate (referred to henceforth as m6A and m6AMP). Thus, the moiety on the 5’ of m6A, either hydroxyl or phosphate, provides a simple and selective way to determine if the m6A derives from mRNA, or contaminating rRNA or snRNA.
The phospho-tag m6A assay is selective only for m6A generated by METTL3/14, not ZCCHC4, METTL5, or METTL16
We first asked if the phospho-tag assay only detects m6A generated by METTL3/METTL14. We therefore prepared total RNA from wild-type mouse embryonic stem (mES) cells as well as Mettl3 knockout mES cells32. We used the Mettl3 knockout mES cells generated by Hanna and colleagues, which exhibit a >99% reduction in m6A levels in highly purified mRNA32,33. Thus, Mettl3 knockout mES cells are an ideal system to establish whether any of the m6A signal in the phospho-tag assay derives from mRNA.
Using the phospho-tag assay, m6A was readily detectable in total RNA prepared from wild-type mES cells (Fig. 1C). In contrast, the m6A signal was reduced by 97.1% in total RNA prepared from Mettl3 knockout ES cells (Fig. 1C). Notably, m6AMP levels were high in both wild-type of Mettl3 knockout samples (Fig. 1C), consistent with the idea that Mettl3 knockout does not affect m6A in rRNA or snRNA. In addition, we treated HEK293T cells for 6 h with 30 μM STM2457, a selective METTL3 inhibitor34. Here we also found that m6A levels were markedly reduced in total RNA (Fig. 1D), further supporting the idea that the phospho-tag assay selectively measures m6A generated by METTL3/METTL14 and not other sources of m6A.
We next confirmed that the m6A signal in the phospho-tag assay is dependent on RNase T1 and does not reflect endogenous 5’-hydroxyl m6A RNAs. Upon removal of the RNase T1 step, which is required to generate m6A with a 5’ hydroxyl, an essentially complete abolition (98%) of the m6A signal was seen (Fig. 1E). Overall, these experiments demonstrate that the RNase T1 step is needed to generate m6A, and that the any detected m6A in the phospho-tag assay was present in a G-m6A context.
We next wanted to establish whether the enzymatic steps had gone to completion. To test this, we used 10 μg RNA from HEK293T cells and compared our RNase T1 treatment (2 h, 2 U enzyme, 37°C) to treatments associated with higher degrees of RNase T1 activity: 4 h reaction times, and adding additional RNase T1 (an additional 2 U added after 2 h, followed by an additional 2 h incubation). In both cases, we found no further increase in m6A levels (Fig. 1F).
We also asked if the nuclease S1 was also optimized. With nuclease S1, the standard reaction condition is 1 h, 10 U enzyme, at 37°C. We tested both extending the reaction time or adding more enzyme. In neither case did we see any increase in the total amount of nucleotide monophosphate levels (Fig. 1G). Based on these results, the RNase T1 and nuclease S1 steps were considered optimized for input RNA levels up to 10 μg.
Overall, these studies demonstrate an assay for selective detection of METTL3/METTL14-derived m6A in total RNA samples despite the presence of vastly larger amounts of rRNA- and snRNA-derived m6A.
A phospho-tag assay for cap m7G to quantify mRNA transcript levels
In the traditional m6A quantification assay, m6A levels are normalized to total adenosine levels in the sample8. However, this approach is problematic since total adenosine levels are influenced by the levels of contaminating rRNA/snRNA, as described above, and the lengths of poly(A) tails. Poly(A) tails is particularly problematic in the m6A field since m6A recruits deadenylases to mediate mRNA degradation35,36. Therefore, adenosine is problematic for normalizing the m6A levels.
Therefore, we decided to develop a new normalization strategy. Rather than normalizing m6A to A, we decided to normalize m6A levels to transcript copy levels. A unique feature of each RNA polymerase II transcript, but not rRNA or snRNA, is the presence of an m7G cap37. This “cap m7G” comprises m7G followed by a triphosphate bridge to the first transcribed nucleotide37. In addition to cap m7G, “internal m7G” is also found in other classes of RNAs, including mRNA, rRNA and tRNA38,39,40 Therefore, the phospho-tag assay needs to distinguish cap m7G from internal m7G.
To distinguish cap m7G from internal m7G, we developed a second approach that again uses phosphate to mark the origin of m7G (Fig. 2A). These steps are designed to occur in the same tube used for m6A quantification above. In this way, parallel sample handling is avoided, thus reducing variability. As a result, m6A and cap m7G can be quantified from a one-pot reaction.
In order to mark cap m7G from internal m7G, we developed a protocol that takes advantage of both RNase T1 and the decapping enzyme yDcpS41. Internal m7G is usually preceded by a nonmethylated G: G-m7G42,43. As a result, after RNase T1 digestion, internal m7G will be released with a 5’-hydroxyl. Therefore, we wanted to mark cap m7G with a 5’-phosphate. We therefore used yDcpS, which releases the cap m7G as a m7GMP41. yDcpS is different from other decapping enzymes such as Dcp2, Nudt12, Nudt15 and RppH which release cap m7G as m7GDP44 (Fig. 2A). We could not use these other decapping enzyme since m7GDP contains a diphosphate, which is not readily detected on mass spectrometry due to ion suppression45. Therefore, yDcpS provides a unique opportunity to release m7G with a phosphate that can reveal that the m7G is derived from the cap.
We first asked if the m7GMP signal derives from the cap. The phospho-tag assay generated readily detectable m7GMP when using total RNA from wild-type mES cells (Fig. 2B). This was primarily derived from cap m7G since omitting yDcpS (2 h, 200 U, 37°C), resulted in a marked reduction (91.1%) of the m7GMP signal (Fig. 2B). The residual non-cap derived m7GMP likely derives from internal m7G nucleotides in cellular RNAs that are not preceded by G38, since these will generate m7GMP after the nuclease S1 step.
Since some m7GMP derives from internal m7G, we included a “no yDcpS” control for each sample in the following experiments. By subtracting this background signal, cap m7G can be more accurately determined.
One potential confounding factor when measuring cap m7G is that some cap m7G derives from short transcripts, such as enhancer RNAs46,47. These short transcripts may have a very low m6A/cap m7G ratio since the transcripts are too short to contain m6A. Since our goal is to measure m6A in mRNA, we used a size fractionation step. In this step, the extracted RNA is subjected to a silica-based purification column that contains wash steps that cause shorter (<200 nt) RNA to be removed.
We asked if the m6A/cap m7G ratio is different in these two samples. In the fraction containing >200 nt RNA, the m6A/cap m7G ratio was ∼25% increased (Fig. 2C), consistent with the idea that the inclusion of small RNAs artificially lowers the m6A/cap m7G ratio. We therefore used the >200 nt fraction for all subsequent experiments.
As before, we determined if the yDcpS treatment was optimized. For these experiments, we again used 10 μg of input RNA after >200 nt size selection from HEK293T cells. Increasing the time or the amount of enzyme did not further increase the yield of m7GMP (Fig. 2D). Thus, the overall phospho-tag assay was considered optimized for both m6A and m7GMP detection for this amount of input RNA.
LC-MS/MS method validation
Compared to other m6A quantification methods48, our phospho-tag m6A assay not only selectively measured mRNA-derived m6A and m7GMP in one sample injection, but also resolved m6A isomers 1-methyladensosine (m1A) and 2’-O-methyladenosine (Am) detection by either LC chromatographic retention time or MS/MS fragment ions (Supplementary Fig. S1). m1A and m6A share the same major fragment ion (quantifier) of methyladenine (M+H, 150.07) and same MRM transition of 282.1→ 150.1, but they can be readily separated by retention time (1.8 vs 2.1 min). While Am is not abundant in our RNA digests, its detection can be distinguished from m6A and m1A by fragment ion of adenine (M+H, 136.06) and transition of 282.1→136.1.
We next validated the phospho-tag m6A assay for method sensitivity, linear range, matrix effects, precision and accuracy. The linearity for the detection of m6A and m7GMP was determined by constructing external calibration curves for unlabeled synthetic standard solution of m6A and m7GMP. To assess the extent of matrix effect on LC-MS/MS measurements, we compared calibration curves prepared from solvent blank (80% methanol, 20% H2O) and matrix solution containing RNA digestion assay buffers (Fig. 3A). In the absence of spiked-in internal standards, both m6A and m7GMP measurements showed fair linearities (R2>0.969) within a dynamic linear range of up to 1μm analytes. The presence of matrix (i.e., RNA digestion buffer) resulted 17% and 26 % decrease in m6A and m7GMP detection sensitivity, respectively.
To eliminate the observed matrix effect and to correct day-to-day, batch-to-batch variability in LC-MS/MS measurement, we spiked in stable isotope-labeled internal standard (SIL-IS), m6A-d3 and m7GMP-d3, into sample matrix at a constant concentration of 1 nM and 5 nM, respectively. As expected, SIL-IS not only eliminated matrix effect, but also improved the linearities of both m6A and m7GMP calibration curves to R2 greater than 0.999 (Fig. 3B). The linear range for both analyte detections were also extended to 2 μm (Fig. 3C). The limit of detection (LOD) and the limit of quantification (LOQ) for m6A and m7GMP calculated as 3 and 10 times blank standard deviation to slope ratio (3*STDEV/slope, 10*STDEV/slope)) were as follows: LOD 0.03 nM and LOQ 0.1 nm for m6A, and LOD of 0.3 nM and LOQ of 1.1 nm for m7GMP.
Several factors can affect the accuracy of measurement including extraction efficiency, stability of the analyte, adequacy of the chromatographic separation and the purity of reference standards49. To test the robustness of phospho-tag m6A assay, we measured the coefficient of variation (CV) in repeated injection of 12 RNA samples and recovery rate of spiked standards to further validate the method precision and accuracy. The recovery rates for 5 nM and 10 Nm m6A and m7GMP standard spiked into RNA digests were in the range of 88-116% which is well within the accepted range of 80-120%. We then assessed the stability of m6A and m7GMP chemical standards spiked into RNA-free matrix and m6A and m7GMP derived from RNA digestion mixture over a time span of 48h. We observed an overall constant readout of ion abundance ratio relative to deuterated internal standard for both analytes over 48h at 4°C, with the coefficient of variations for RNA sample and chemical standard well within 9.7% (Fig. 3D). Note that the total LC-MS/MS sample run is 8 min, phospho-tag can easily achieve >300 sample run with accuracy and robustness.
Input requirements for the phospho-tag m6A assay
We next determined the minimal RNA input requirements needed to quantify m7GMP and m6A. Our optimizations for RNase T1 and yDcpS used 10 μg, but it would be desirable to know the minimum amount of input RNA that would provide accurate m6A and cap m7G measurements. We tested 1, 2.5, 5, 7.5 and 10 μg amounts of input RNA using the phospho-tag assay in two biological replicates. At all these RNA sample inputs, we were readily able to measure m6A and m7GMP (Fig. 4A). Over this range of RNA input, the m6A and m7G levels showed linearity (correlation coefficients r2□> □ 0.995). These data suggest that m6A and m7GMP measured using RNA input levels as low as 1 μg can be reliably measured in this assay.
Notably, the limiting factor for accurate quantification of the m6A:m7GMP ratio in the phospho-tag m6A assay was the detection of m7GMP. The m6A measurements were linearly increased with increase in the RNA input from 1 to10 μg (r2□> □ 0.995) (Fig. 4A).
However, the m7GMP measurements didn’t display the degree of linearity with increase in RNA input as compared to m6A (r2□ = □0.91). The linearity of m7GMP was markedly improved (r2□ = 0.98) by use of m7GMP-d3 spike in at constant concentration of 5 nM in each sample post methanol precipitation step. The poor detection of m7GMP with low input RNA likely reflects the reduced ionization efficiency of m7GMP, a phenomenon seen with other phosphorylated molecules45.
We next assessed the quantitative accuracy of the phospho-tag m6A assay. To do this, we prepared total RNA from wild-type and Mettl3 knockout mES cells. We then mixed these RNAs in specific ratios and estimated the quantified the m6A:m7GMP ratio. For all experiments, the sum of wild-type and Mettl3 knockout RNA was 10 μg. Here we found that the m6A:m7GMP using the phospho-tag m6A assay directly correlated with the fraction of wildtype RNA mixed with the Mettl3 knockout RNA (Fig. 4B). Overall, these studies suggest that the phospho-tag m6A assay shows high quantitative accuracy.
Quantitative analysis of m6A/m7GMP ratios using stable isotopically labeled-internal standards (SIL-IS)
To increase the quantitative accuracy of m6A and m7GMP detection, we analyzed samples with stable isotopically labeled-internal standard (SIL-IS) of m6A and m7GMP. Use of SIL-IS improves reproducibility between injections, compensates for the loss of sensitivity during a batch run of samples, and accounts for matrix effects that can happen during the ionization process50,51.
To test this, we prepared input RNA from HEK293T cells and measured the m6A:m7GMP ratio in three biological replicates in which we either did not add standards or added standards. For the samples with standards, we mixed the m7GMP-d3 and m6A-d3 standards after the methanol precipitation step at a concentration of 5 nM and 1 nM, respectively. We then calculated the m6A:m7GMP ratio for both sets of samples. In the case of the samples without standards the average m6A:m7GMP ratio was 20.3 +/- 10.3 % RSD, while the average m6A:m7GMP ratio was 1.7 +/- 3.9% when standards were used (Fig. 4C). These results show that the variability in measurement is markedly reduced when using the SIL-IS standards.
An additional benefit of using SIL-IS is that the m6A:m7GMP ratio can be determined using exact amounts of m6A and m7GMP. Thus, results with SIL-IS also reveal the exact ratio of m6A per transcript. For all subsequent experiments SIL-IS was used to quantify the m6A:m7GMP ratio.
DISCUSSION
A major problem with m6A assays is that they can be highly affected by the way in which the sample is prepared. Most assays rely on poly(A) purification, but this method typically is associated with residual contaminating rRNA. Additionally, poly(A) purification is associated with preferential accumulation of mRNA from 3’ ends, a region which is known to be enriched in m6A. Since these problems are hard to control between samples, sample-to-sample variation in m6A levels, as measured by mass spectrometry or other methods, may simply reflect variation in these sources of error. The phospho-tag m6A assay overcomes this problem by selectively measuring only m6A from mRNA. Thus any rRNA contamination does not contribute to the overall m6A measurement. Additionally, since total mRNA is used, there is no bias for 3’ ends of mRNA. As a result, all parts of the mRNA are used in this assay. Notably, the phospho-tag m6A assay also includes normalization to m7G caps, thus providing insights into the overall level of m6A per transcript.
The core concept of the phospho-tag m6A assay is that phosphates are used as an endogenously derived mass tag to indicate the origin of m6A. m6A from rRNA or snRNA will produce m6A with a 5’ phosphate. In contrast, m6A derived from mRNA will produce m6A as a nucleoside, i.e., with no phosphate, as long as the m6A was preceded by G. Importantly, ∼75% of all m6A in mRNA is preceded by a G21,22,26,52,53. Thus, the amount of m6A calculated per transcript can be adjusted to take into account that the value calculated in the phospho-tag assay is ∼75% of the total level of m6A in mRNAs. The presence or absence of phosphate changes the mobility of m6A in the LC step of the MS analysis and provides a unique mass. In this way, m6AMP cannot be accidentally measured and used in the calculation of the overall m6A level in mRNA.
The phosphate tag on m6A derived from rRNA and snRNA, but not mRNA, is a result of the selectivity of RNase T1. RNase T1 cleaves RNA after G residues, and importantly, leaves the phosphate on the 3’ end of the G. As a result, the subsequent nucleotide has a 5’ hydroxyl. Since m6A in mRNA typically is preceded by a G, it will contain a 5’ hydroxyl after RNase T1 treatment. In contrast, the A-m6A bond in rRNA and the C-m6A bond in snRNA will not be cleaved by RNase T1. Instead, these bonds are cleaved by nuclease S1 in the subsequent step. Nuclease S1 cleaves the RNA, leaving the phosphate on the m6A. In this way, the phosphate acts as a tag that reveals the RNA origin of the m6A.
The second core idea of the phospho-tag assay is that m6A levels are normalized to mRNA abundance by using the m7G cap as a surrogate for the overall mRNA transcript levels. In previous studies, m6A was normalized to total A levels in a sample. This is problematic because rRNA is typically present in mRNA samples, and thus, this normalization is highly affected by the level of rRNA contamination. In contrast, since rRNA and snRNA lack m7G caps, their presence does not affect the overall quantification of m6A/m7GMP.
Since m6A is normalized to transcript levels, it is important to keep in mind that changes in m6A/m7GMP levels might reflect changes in mRNA 3’UTR lengths. In some conditions, 3’UTR lengths change as a result of regulated polyadenylation site selection. Thus, if an mRNA becomes longer due to a longer 3’UTR, it may have more m6A per transcript. However, this does not mean that the stoichiometry of m6A sites has increased. Thus, any observed changes in m6A levels between two samples should be orthogonally validated using site-specific measurements of m6A stoichiometry using assays such as SCARLET54.
The cap-derived m7G is also selectively marked by a phosphate. This is achieved using yDcpS, an enzyme the selectively hydrolyzes m7G that is part of the mRNA cap. yDcpS releases m7G with a 5’ monophosphate. Importantly, this reaction is performed along with the RNase T1 reaction in a single pot. As a result, internal m7G nucleotides, which are typically found in rRNA and tRNA preceded by a G, are released from RNA as a m7G with a 5’ hydroxyl. In this way, the cap-derived m7G can be selectively measured based on its unique mass and retention time conferred by its 5’ phosphate.
Notably, there is a small amount of m7G in cellular RNA that is not preceded by a G. This is seen in samples that are not treated with yDcpS. Therefore, a “no yDcpS” control should be used to establish this background level of m7GMP. In contrast, very little m6A containing a 5’ hydroxyl is generated in samples where RNase T1 was omitted. Nevertheless, a “no RNase T1” control is important to ensure that the m6A signal indeed derives from cellular mRNA.
The phospho-tag m6A assay was validated using a Mettl3 knockout embryonic stem cell line. Importantly, total RNA was used in these experiments. Thus, sample preparation was highly simplified. Despite the large amount of m6A from rRNA in these samples, no m6A was detected since this assay is highly selective for m6A derived from mRNA.
Overall, we expect that the phospho-tag m6A assay will allow rapid and simple measurements of m6A from essentially and RNA sample without the need for tedious poly(A) purification. Also, since this reaction is a single-pot reaction, it can be used in medium- and high-throughput assays for m6A measurements. We expect that the phospho-tag m6A assay will reveal whether m6A is dynamic and will help to identify the specific signaling or disease contexts in which these dynamics occur.
MATERIALS AND METHODS
rRNA mapping
Publicly available raw RNA-Seq and m6A-Seq reads from Gene Expression Omnibus database (GEO) were downloaded from two studies (GEO accession: GSE144984 and GSE113798). The fastq files quality was checked with MultiQC tool55. Adapters were removed using Trimmomatic56. The adapter-trimmed reads were mapped to the human rDNA complete unit (KY962518.1) using Bowtie257 (version 2.4.2) with --local option. For each sample number of mapped and unmapped reads were used to calculate the percentage of rRNA reads. The rRNA reads from biological or technical replicates were averaged.
Cell line culture
HEK293T/17 and mouse embryonic fibroblast cells (mEFCs) were cultured in 1× DMEM (Life Technologies #11995-065) with 10% FBS, 100□U□ml−1 penicillin and 100□μg□ml−1 of streptomycin under standard tissue culture conditions. Cells were split using TrypLE Express (Life Technologies) according to manufacturer’s instructions. Cells were harvested after reaching 80% confluency. Mouse embryonic stem (mESCs) were previously described by Geula et al (32),and were a kind gift from S. Geula and J.H. Hanna (Weizmann Institute of Science). All mESCs were grown in tissue culture plates precoated with 0.1% gelatin (EmbryoMax ES-006-B) in mESC media (KnockOut DMEM (Gibco #10829018), 15% heat-inactivated fetal bovine serum (FBS) (Gibco #26140079), 100 μg/ml streptomycin (Gibco #15140122), 100 U/ml penicillin, 1x GlutaMax (Gibco #35050061), 55 μM β-mercaptoethanol (Gibco #21985023), 1x MEM non-essential amino acids (Gibco #11140076), 3 μM CHIR99201 (Sigma Aldrich SML1046), 1 μM PD0325901 (APExBIO # A3013), 1000 U/ml LIF (Millipore # ESG1107).
RNA isolation
RNA was isolated using TRIzol™ LS Reagent (ThermoFisher #10296010) following manufacturer’s instructions except after adding chloroform the sample was transferred to pre-spined MaXtract High Density (Qiagen #129065) and centrifuged at 12,000g for 15 mins at 4°C to phase sperate the aqueous phase from the organic phase. Next the aqueous phase containing RNA was transferred into a fresh tube and standard protocol was followed for total RNA isolation. Next, the isolated RNA was treated with DNase I (Invitrogen™ #18047019) at 37°C for 1 hour to remove traces of contaminating DNA. The DNase I treated RNA was further purified into small (<200nt) and large (>200nt) RNA fractions using RNA Clean & Concentrator™-25 (Zymo Research # R1017) following the manufacturer’s instructions. The large (>200nt) RNA fraction was used as the input for Phospho-tag assay.
Enzymatic digestion of RNA
10 μg of >200nt RNA (in nuclease-free water) was decapped using 200 U yDcpS and yDcpS buffer (NEB, #M0463S) at 37°C in a thermomixer at 800rpm pulse shaking. Next 2 U of RNase T1 (INVITROGEN #AM2283) was added to the tube and incubated at 37°C in a thermomixer at 800rpm pulse shaking. To the above tube,10 U of S1 Nuclease (INVITROGEN #18001-016) was added, and sample was incubated for 1 hour at 37°C in a thermomixer at 800 rpm pulse shaking to digest RNA to mononucleotides. Next 4 volumes of 100% methanol was added to facilitate precipitation of all enzymes present in the hydrolysate before the sample injection. The samples were centrifuged at 16,000g for 30 min at 4°C and supernatant was carefully transferred into a fresh 1.5 mL tube without disturbing the pellet. Additionally, for each phospho-tag m6A assay two control reactions were step up: (i) yDcpS is omitted (ii) RNase T1 is omitted. The signals from these two controls are subtracted from the sample in which all enzymes were added.
Preparation of calibration with and without SIL-IS
For preparation of calibration solutions synthetic standard of N6-methyladenosine (m6A) was purchased from Selleck Chemicals, Houston, Texas, USA (Catalog No. S3190) and 7-Methyl-guanosine-5’-monophosphate (m7GMP) was purchased from Jena Bioscience, Jena, Germany (Catalog No. NU-1135S). N1-methyadenosine (m1A) was purchased from (Carbosynth #NM03697). The deuterated standards m6A-d3 and m7GMP-d3 were custom synthesized from Toronto Research Chemicals, Ontario, Canada).
Synthetic standards of m6A and m7GMP were dissolved in pure water at a final concentration of 1 mM each. From these stocks two set of calibration solutions were prepared in the range of 0 nM to 1 μM in sample matrix and 80% methanol. The sample matrix consists of all the reaction and enzyme storage buffers in 80% methanol. Additionally, calibration sets were also spiked with SIL-IS, m6A-d3 and m7GMP-d3 at constant a final concentration of 1 nM and 5 nM throughout all calibration samples, respectively to correct for sample matrix effect. For each Phospho-tag assay, 50 μl of methanol precipitated hydrolysate was aliquoted into HPLC vials and m7GMP-d3 and m6A-d3 was spiked in at a final concentration of 5 nM and 1 nM, respectively.
Stability test for m6A and m7GMP
To assess the stability of m6A and m7GMP we performed repeated injections of two set of samples spiked with synthetic m6A, m7GMP and SIL-IS m6A-d3 and m7GMP-d3 at 10 nM over a period of 48 hours in sample matrix and 80% methanol. We performed another set of repeated injections of digested RNA(>200nt) spiked with SIL-IS at 10 nM over 48 hour period. We then calculated % RSD (relative standard deviation) for all three sample sets. The samples were maintained at 4°C.
Procedure for LC-MS/MS measurements
RNA digests were analyzed by LC-MS/MS using a platform comprised of an Agilent Model 1290 Infinity II liquid chromatography system coupled to an Agilent 6460 Triple Quadrupole mass spectrometer equipped with Agilent Jet Stream Technology. Chromatography of metabolites utilized reversed-phase chromatography on a Infinity Lab Poroshell 120 EC-C18 column (Agilent Part Number:695975-902). Mobile phases consisted of: (A) 99% water, 1% acetonitrile containing 1 mM ammonium formate and 0.1% formic acid, and (B) 99% acetonitrile, 1% water containing 1 mM ammonium formate and 0.1% formic acid. The following gradient was applied at the flow rate of 0.9 ml/min: 0-0.5 min, 90% A,10% B; 0.5-1.0 min, 70% A, 30% B; 1-4 min, 30%, 70% B; 4-4.1 min, 1% A, 99% B, to 6 min, followed by a re-equilibration at 90% A for 2min. The column compartment temperature was at 25°C. The injection volume is 2 μl. MRM data were acquired in positive ion mode. Fragmentor, collision energy, and other source parameters were optimized for both quantifier and qualifier ions using Agilent MassHunter Optimizer software (version 6.0). Source parameters for m6A measurement were as follows: gas temperature, 230°C; gas flow, 8 l/min; nebulizer, 35 psi; sheath gas temperature, 400°C; sheath gas flow, 12 l/min; capillary voltage, 2500v; delta EMV, 400 v. The source parameters for m7GMP are the same as m6A. The quantification of m6A and m7GMP were achieved using MRM transitions for both quantifier and qualifier ions shown in Table 1.
The LC-MS/MS data was analyzed using Agilent MassHunter Quantitative Analysis (for QQQ).
AUTHOR CONTRIBUTIONS
S.R.J. and A.H.M. conceived and designed the experiments. A.H.M. carried out experiments, analyzed data, and prepared figures. Q.C. developed the LC-MS/MS part of the assay and N.A. performed LC-MS/MS measurements. S.R.J. and A.H.M. wrote the manuscript with help from all the authors.
Competing interests
S.R.J. is scientific founder of, is advisor to, and owns equity in Gotham Therapeutics and 858 Therapeutics.
ACKNOWLEDGEMENTS
We thank members of the Jaffrey Lab for helpful comments and suggestions. This work was supported by NIH grant R35NS111631 to S.R.J, and T32CA062948 to A.H.M.