High Throughput and Low Bias DNA Methylation and Hydroxymethylation Analysis by Direct Injection Mass Spectrometry

Abstract

We present a direct injection mass spectrometry (DI-MS) platform that accurately, precisely, and quickly quantitates global levels of DNA cytidine methylation (5mC) and hydroxymethylation (5hmC). Our platform combines an Advion TriVersa NanoMate coupled online to a Thermo Scientific Orbitrap Fusion Lumos. Following digestion to nucleosides, the DNA samples are analyzed at the rate of <1 min per injection with comparable detection limits of 0.63 ng/μL and 0.31 ng/μL, respectively. In contrast, the detection limits for 5mC and 5hmC in state-of-art nano liquid chromatography (LC) coupled to online mass spectrometry (nLC-MS) are notably different (0.04 ng/μL and 2.5 ng/μL, respectively). The high sensitivity of DI-MS is achieved by optimizing sample buffer composition, the source fragmentation energy, and the radio frequency of the instrument ion funnel. DI-MS accurately reports the relative abundance of 5mC and 5hmC over a range of 1% to 7% (R² > 0.98) and 0.13% to 1.75% (R² > 0.99), respectively. Accurate measurement of C, 5mC and 5hmC is achieved by optimizing in-source fragmentation to obtain a population of up to 93% of just the nucleoside base. This protocol minimizes base dimer formation and partial base-deoxyribose dissociation in gas phase and greatly improves modified base quantitation. We also demonstrate that DI-MS overcomes biases in differential chromatographic retention and issues of sample degradation in the autosampler due to its high throughput. Finally, we present an application of our workflow to quantify DNA modifications on a batch of 81 samples in about 1.5 hours.

Graphical Abstract

INTRODUCTION

Epigenetic regulation and disease related misregulation

Methylation of cytosine (5mC) is an epigenetics marker that regulates gene expression.^1-4 Gene silencing due to 5mC formation is associated with heterochromatin formation via coordinate action of the DNA methylation and histone modifying enzymes.⁵ For example, the regulatory protein MeCP2 binds islands of 5mC and silences gene transcription by recruiting histone deacetylase or histone methyltransferase; the latter enzyme adds the H3K9 methylation mark.^5-8 The DNA methyltransferases (DNMTs) repress gene expression by interacting with histone methyltransferase or histone deacetylases (HDACs).^{5, 6, 9} Islands of 5mC are spread across the genome comprising 70% of the 2.8 x 10⁷ CpG sites. The exceptions are CpG islands (500-1000 bp) that contain the regions of DNA where ‘housekeeping’ genes are found.^{6, 10} Hypermethylation of intergenic regions represses expression of transposable and viral elements incorporated in mammalian genome.^{6, 7, 10}

A repertoire of enzymes adds, removes, and recognizes sites and clusters of 5mC. Proteins that specifically bind 5mC and repress transcription include the aforementioned MeCP2, the family of Methyl-CpG Binding Domain (MBD) proteins, UHRFs, Kaiso and ZBTBs.^{7, 11, 12} The DNMTs catalyze the transfer of a methyl group from S-adenyl methionine onto the C-5 position of cytosine (5mC).⁶ De novo DNA methylation is catalyzed by DNMT3a and DNMT3b. DNMT1 catalyzes the methylation of the progeny DNA to match that of the parental during replication.^{2, 7} DNA methylation has as opposing mechanism, demethylation, which is normally achieved by converting an amine into a carbonyl group. Specifically, the AID/APOBEC enzyme converts 5mC directly into thymine or a cytosine into uracil.^{13, 14} Alternatively, 5mC is oxidized by ‘ten-eleven translocation enzymes’ (TET) into 5-hydroxy methyl cytosine (5hmC).¹⁵ 5hmC is converted to 5-formyl cytosine (5fC) by TET2 and into carboxy cytosine (5caC) by TET3.^16-18 The base excision repair pathway (BER) removes these modified residues yielding unmodified cytosine.^{6, 7, 17} The latter removal steps are slow relative to TET1 activity resulting in detectable 5hmC in genomic DNA.

Hypomethylation of cancer cells within intergenic and oncogene regions promotes anomalous gene transcription and expression. Conversely, tumor suppressor genes are hypermethylated thereby repressing their expression.¹⁰ Genomic instability as a result of global hypomethylation is observed in tumor or cancer cells. Altered DNA methylation is also observed in non-cancer related disease states like neurological diseases.^{6, 19} While DNA methylation and alteration in its patterns is a complex process, in-depth studies are facilitated by techniques that can detect and quantify low abundant modified forms of cytosine notably, 5mC and 5hmC. Quantitation of the amount of these modifications is critical to understanding mechanisms of epigenetic regulation and provides diagnostic markers for disease states such as in cancer.¹⁰ More than 50% of human cancers have mutations in factors involved in chromatin compaction, and cancer cells routinely use epigenetic processes to escape the immune system and chemotherapy.²⁰ DNA methyltransferases were the first epigenetics target for cancer treatment, with the development of the DNMT inhibitor 5-aza-2′-deoxycytidine, now sold under the brand name Vidaza. Since then, numerous epigenetic drugs have been FDA approved for cancer therapy, including multiple inhibitors of the DNA methyltransferase (DNMTs) or demethylase (TET) process.

Quantitation of 5mC and 5hmC present in genomic DNA by mass spectrometry

Next-Generation Sequencing of bisulfite-converted DNA enables genome-wide localization of islands of DNA methylation.²¹ When only amount rather than sequence location is required, levels of DNA methylation can be assessed rapidly and cost-effectively by mass spectrometric analysis of mononucleosides derived from chemically or enzymatically digested DNA.^{22, 23} The presently used method of liquid chromatography tandem mass spectrometry (LC-MS/MS) can quantify methylation levels from 0.05-10% and detect as little as ~ 0.25% modified cytosine from 50-100 ng of DNA. This detection level is the equivalent of 5% differences in global levels of 5mC, 5hmC, 5fC and 5caC.^{18, 22, 24-26} A number of LC-MS/MS methods have been published that reflect the need to optimize chromatographic separation due to the hydrophilic nature of single nucleosides.^{27, 28}

Optimizing LC analysis of nucleosides includes optimizing column retention of the analyte, minimizing sample carryover, controlling for batch effects between experiments due to the use of different columns or different calibration, and accurate quantitation of peak areas in the extracted ion chromatogram. Multiple Reaction Monitoring (MRM) is most frequently used for nucleoside analysis in mass spectrometry, which consists in acquiring only a selected list of signals defined by the user.^{24, 25, 29, 30, 31} Targeted approaches to LC/MS such as MRM improve throughput and more sensitive since the mass spectrometer does not occupy duty cycle time in selecting unknown analytes. However, targeted methods such as MRM preclude the discovery of new molecular species. In addition, MRM can yield inaccurate results due to nucleoside dimer formation or source fragmentation that is not accounted for in the analysis.^{25, 32}

The direct injection mass spectrometry (DI-MS) platform for 5mC and 5hmC quantitation we present herein is orders of magnitude faster than LC, accurate, precise, and untargeted. DI-MS leverages the capability of the ultra-high resolution Thermo Scientific Orbitrap Fusion Lumos mass spectrometer to analyze nucleosides at the MS1 level with high mass accuracy and reproducibility. We have demonstrated that the Advion TriVersa NanoMate automated ion source used in DI-MS operates quickly and eliminates sample cross-contamination.³³ Our workflow optimizes sample digestion and desalting for DI-MS. Our method enables large-sample-number high-throughput applications and bypasses common LC-MS issues such as sample carryover, lack of sensitivity due to poor binding of nucleosides to the column, and imperfect resolution of the peaks by chromatography. The result is a new method that is fast, cost-effective, and accurate.

EXPERIMENTAL SECTION

Materials for sample preparation

Calibration Standards:

A set of three 897 bp dsDNA standards (Zymo Research) containing either unmodified-cytosine, 5-methylcytosine (5mC) or 5-hydroxymethylcytosine (5hmC) are used for calibration curve and detection limit analyses. The unmodified standard contains 100% cytosine (C) in the DNA sequence. The methylated standard contains 100% 5mC and the hydroxymethylated standard contains 100% 5hmC. The 5mC calibration curve is prepared by mixing 50 ng of the unmodified standard with 0.5, 1.0, 1.5, 2.5 and 4 ng of the 5mC standard yielding 1, 2, 2.9, 4.7 and 7.4% of 5mC respectively. The 5hmC calibration curve is prepared by mixing 50 ng of the unmodified standard with 0.0625, 0.125, 0.25, 0.5 and 1 ng of 5hmC standard yielding 0.125, 0.25, 0.5, 1 and 1.96% of 5hmC, respectively. Standard solutions for a mixture of modified cytosines is prepared by mixing 100 ng of the unmodified standard with the amounts noted above of the 5mC and 5hmC yielding 1, 2, 3, 5 and 7% 5mC and 0.13, 0.25, 0.5, 1 and 1.75% 5hmC. These solutions are digested with the Nucleoside Digestion Mix (New England BioLabs), desalted with graphite, dried in a vacuum centrifuge and resuspended in 70% acetonitrile (ACN).

Cell culture:

The cell lines HepG2/C3A, THLE-3, HFF-1, Daoy and HEY were obtained from the American Type Culture Collection (ATCC). The cell lines HeLa, 293F, 293T, HEK Freestyle, Expi293, ExpiCHO, S2 and SF9 were kindly provided by the Macromolecular Therapeutics Development Facility of the Albert Einstein College of Medicine, New York, USA. More information for the cell lines are provided in the Supporting Information.

DNA extraction, hydrolysis, and RNA removal:

DNA from the 1x10⁶ cells of each cell line noted above was extracted using the DNeasy Blood and Tissue kit (Qiagen), according to the manufacturer’s instruction. RNA-free genomic DNA was obtained by incubating samples for 2 min, at room temperature, with 4 μL of RNase A (100 mg/mL). Column washes according to kit recommendation removed the digested RNA. Eluted DNA was assessed for concentration (A₂₆₀) on a NanoDrop ND1000 instrument (Thermo Fisher). Following extraction, DNA hydrolysis was performed by using Nucleoside Digestion Mix (New England BioLabs). Briefly, 0.5 μg of DNA was mixed with 2 μL of reaction buffer, 1 μL of nucleoside digestion mix and water to a final volume of 20 μL.

Sample desalting for DI-MS:

All samples described in the manuscript were desalted before being direct injected into the MS using a homemade filter composed of a plug of C18 filter pad (Empore) stuffed into a 200 μL pipet tip that was loaded with 60 μL (~1 μg) poly-graphitic carbon resin (PGC, HyperCarb, Thermo Scientific). The filter was washed with 100 μL 0.1% TFA prior to use. Digested DNA sample was mixed with 100 μL 0.1% TFA (pH = 2~3), loaded onto the filter and washed once with 100 μL 0.1% TFA. Nucleosides were eluted with 60 μL of a buffer composed as 60% acetonitrile (ACN) and 0.1% formic acid (FA) and dried in a vacuum centrifuge. For large amount of samples, commercially prepacked PGC 96 well plate is also available. The recovery from PGC nucleoside desalting was tested using extracted DNA from HepG2/C3A cells followed by digestion. The DNA concentration (A₂₆₀ present in each preparation was assessed using a NanoDrop ND1000 instrument (Thermo Fisher) before and after desalting, confirming a recovery rate over 92%.

Nucleoside stability test

The nucleosides 2’- deoxyguanosine monohydrate (98+%, Alfa Aesar), 2’- deoxycytidine (99+%, Acros Organics), 2’- deoxyadenosine (99%, Alfa Aesar), Thymidine (99+%, Acros Organics) and Cytidine (Aldrich) were each diluted in 0.1% TFA to a final concentration of 10 mM. One μl of a 30 μM equimolar mixture of these nucleosides was aliquoted in each well of a 96-well plate; the plate was then dried using a vacuum centrifuge. Samples were injected as follows: for the aliquots in solution, the autosampler picked up 5 μL (15 μM) of sample volume and loaded it onto the two-column system; for the dried sample, the autosampler was programmed to collect 10 μL of 0.1% TFA from a prep vial, resuspended the sample in the well and picked up 5 μL (15 μM) for loading. Samples were assayed hourly for 12 hours every 30 minutes and then daily for 14 days.

Nano liquid chromatography coupled to mass spectrometry (nLC-MS) instrument settings

nLC was configured with a two-column system consisting of a 300 μm ID x 0.5 cm C18 trap column (Dionex) and a 75 μm ID x 25 cm Reprosil-Pur C18-AQ (3 μm; Dr. Maisch GmbH, Germany) analytical nano-column packed in-house using a Dionex RSLC Ultimate 3000 (Thermo Scientific, San Jose, CA, USA). nLC was coupled online to an Orbitrap Fusion Lumos mass spectrometer (Thermo Scientific). The spray voltage was set to 2.3 kV and the temperature of the heated capillary was set to 275 °C. The full scan range was 110–600 m/z acquired in the Orbitrap at a resolution 120,000. The source fragmentation energy was set at 30 V and RF lens % was set at 50. Extracted signals are the protonated nucleobases Cytosine (C) and methylcytosine (mC) with m/z at 112.0505 and 126.0662, respectively.

Direct injection coupled to mass spectrometry (DI-MS) instrument settings

DI-MS was performed with a TriVersa NanoMate (Advion) coupled online with the Orbitrap Fusion Lumos mass spectrometer (Thermo Scientific). The NanoMate was programmed to pick up 5 μL of solution followed by 0.5 μL of air gap to avoid spilling. Samples were sprayed into the mass spectrometer using a gas pressure of 0.3 psi and a positive voltage set at 1.7 kV. No flowrate could be set or easily calculated, as it was contingent to the gas pressure, the voltage and the buffer composition, although it was estimated to be between 100-300 nL/min (personal communication with NanoMate engineer). Contact closure to start MS acquisition was set at 2.5 seconds after engaging the probe to the instrument chip nozzle. MS settings were the same as nLC-MS.

Data analysis and availability

Raw data files were manually analyzed using the software Xcalibur (Thermo Scientific). For nLC-MS runs, the abundance of the nucleosides was calculated by performing extracted ion chromatography and integrating the area below the chromatographic curve. For DI-MS, the abundance was calculated by extracting the intensity of the m/z value in single mass spectra. The reproducibility of the analysis was assessed by using relative standard deviation. All raw files are available in the Chorus repository (https://chorusproject.org) under the project number 1718.

RESULTS AND DISCUSSION

To assess the differences and potentially the advantages in performing modified nucleoside analysis via DI-MS vs LC-MS, we proceeded as follows: (i) first, we identified the type of signals detected in mass spectrometry after electrospray ionization, which led to the identification of numerous byproducts including dimers and salt adducts. (ii) Next, we compared the reproducibility of the analysis via LC-MS and DI-MS. (iii) Next, we performed a series of optimizations of MS settings to improve sensitivity and reduce formation of nucleoside byproducts suitable for both LC-MS and DI-MS detection, (iv) followed by optimization of buffer composition only applicable to DI-MS detection. (v) We then evaluated the potential carryover of LC-MS injections and (vi) assess the stability of nucleosides in solution, which is a potential issue for low throughput analyses. (vii) Correction from observed value to theoretical was achieved by performing calibration curves of 5mC and 5hmC for both DI- and LC-MS methods. (viii) We also measured the biases in nucleoside quantification for the two methods, and we identified to be mainly derived by the poor LC retention of 5hmC using the chromatographic approach and (ix) inevitable incomplete digestion of DNA affecting mostly LC-MS detection compared to DI-MS. (x) Lastly, is presented a methylation level comparison between LC- and DI-MS using DNA extracted from different cell lines. We conclude that DI-MS has minimally less sensitivity compared to LC-MS, but has a significant advantage in throughput and unbiased quantification of modified nucleosides.

Dimer and salt adduct byproducts

We first conducted an LC-MS analysis of an equimolar mixture of synthetic deoxynucleosides, dA, dG, dC and T. We focused first on deoxycytidine given our interest in quantifying the relative abundance of its modifications. Full MS scans revealed that deoxycytidine (dC) can fragment in source to the base (C), form dimers in the gas phase (dC-dC), and also form salt adducts (dC-Na and dC-dC-Na) despite samples being desalted during trap column loading (Figure 1A, 1B). Chen et al., ³⁴ reported the formation of salt adducts which indicates it is a common phenomenon in mass spectrometry detection. Base dissociation from the deoxyribose and dimer formation occurs in the gas phase. The different species are not chromatographically separated indicating that only a single type of each molecule binds to the chromatographic column (Figure 1C).

Figure 1 – Detection and quantification of cytosine forms.

(A) Predicted forms in gas phase of the cytosine based on detected spectra. From left to right: deoxycytidine (dC), cytosine (C), sodium adduct (dC-NA), dimer (dC-dC) and dimer with sodium adduct (dC-dC-Na). (B) Full MS spectrum of injected C, showing the relative abundance of the five products with corresponding m/z. (C) Extracted chromatogram of the five products with the same elution time, indicating formation in gas phase rather than solution. (D) Extracted chromatogram of thymidine (T), deoxyadenosine (dA) and deoxyguanosine (dG) with different elution time. (E) Area of extracted ion chromatography vs signal intensity of C detected by LC-MS (left) and DI-MS (right) respectively using the digested 897 bp dsDNA standard. Six individual replicates were aquired for each method, and the variation was assessed using relative standard deviation. The y-axis is expressed as Log10 intensity.

To confirm that the co-elution of the abovementioned molecules is not just due to poor chromatographic resolution, we extracted the m/z values of the other nucleosides (T, dA and dG) proving baseline separation and resolution is good (Figure 1D). But, we cannot rule out the possibility that the salt adduct formation may occur in solution. Nevertheless, the potential differences in the ionization efficiencies of the analytes, and the loss in sensitivity caused by splitting the total dC signal into at least 5 different species, demonstrates that acquisition optimization is necessary for minimizing quantification biases.

Optimizing MS detection of nucleosides

Both LC and DI methods of nucleoside analysis benefit from optimizing the detection of modified nucleosides. To generate a homogenous population of ions, we and other investigators³⁴ apply CAD (collisionally activated dissociation) fragmentation on the instrument source to fragment all molecular forms into single nucleobases. On our Orbitrap Fusion Lumos, fragment dC, dC-Na, dC-dC and dC-dC-Na into nucleobase C. Using LC-MS to quantitate the species present, we optimized the source fragmentation energy spanning from 5 to 70 V (Figure 2A). The population of cytosine nucleobase C (green) increases up to a maximum of 93% at 35V. Above 35V, the fraction of C dips sharply. This dip is not due to the more efficient production of other species, but rather an overall reduction in signal transmission efficiency (Figure 2B). Therefore, we utilized 30V as default source fragmentation energy for the rest of the analyses presented herein. This voltage is the optimum compromise between sensitivity and minimal presence of nucleoside byproducts.

Figure 2 – Optimization of the MS source fragmentation energy, MS radio frequency (RF) voltages and sample buffer composition.

(A) Relative population of the five products C (blue), dC (orange), dC-Na (gray), dC-dC (yellow) and dC-dC-Na (green) as function of the source energ spanning from 5 to 70 V. The total population is normalized to 100%. (B) Abundance of the five products changing as function of the source energy. (C) Normalized population of the five products as function of the instrument radio frequency (RF) lens %. (D) Absolute intensities of the five products as function of the RF lens %. (E) Signal intensity of C after re-suspending pure dC (45 μM) in different concentrations of ACN + 0.1% FA, (F) different concentrations of methanol + 0.1% FA, (G) different concentrations of FA + 30% ACN and (H) different concentrations of ACN.

To maximize the efficient capture and focus of the ions at the entrance of the mass spectrometer, we also optimized the radio frequency (RF) of the ion funnel at the entrance of the Orbitrap Fusion Lumos. This optimization was performed as different m/z signals have different optimal RF values. The ion funnel is a series of concentric electrodes to which RF voltages are applied, with opposite phases applied to even- and odd-numbered electrodes that generates confining RF electric fields to focus the ion beam. For this optimization, the source fragmentation energy was preset to 30V as per the already standardized protocol. Although there is a slight decrease in the fraction of C population in proportion to the other byproducts when the RF lens increases from 20 to 80%, the overall fraction of C remains over 90% (Figure 2C). The abundance of C was highest at 50% RF lens value, and thus 50% was the setting utilized in our following experiments.

Optimizing sample solvation for DI-MS

We then connected the Advion TriVersa NanoMate to the mass spectrometer in order to optimize solvation for direct sample injection. In LC-MS, the solution in which the sample is sprayed into the mass spectrometer is normally limited to the buffer utilized for chromatographic separation. Since DI-MS is more flexible, we systematically tested different organic solvents and acid concentrations in order to maximize the MS signal. We evaluated the signal intensity of C by re-suspending pure dC (45 μM) in different concentrations of ACN + 0.1% FA or different concentrations of methanol + 0.1% FA (Figure 2E-F). The MS settings optimized for LC-MS described above were used for these DI-MS studies. Seventy percent organic solvent provides the most intense DI-MS signal of C. This high concentration of organic solvent cannot be utilized in LC-MS as nucleosides elute from a C18 column in the 1-30% ACN concentration range (data not shown).

While both ACN and methanol provided very comparable sensitivities (Figure 2E-F), we preferred ACN for the experiments that follow. We next tested the ideal concentration of FA in the signal using 30% ACN as a reference. This percentage, rather than the sensitivity apex of 70%, enables the impact of FA on signal intensity to be efficiently monitored. We observed that in DI-MS, the higher the concentration of FA in solution, the lower the signal intensity of C (Figure 2G). Use of FA in LC-MS to facilitate analyte retention in C18 chromatography thus requires a trade off in reduced signal intensity. As confirmation, we reproduced the data from Figure 2F without the use of FA, and observed that the maximum sensitivity is achieved at 70-80% ACN in solution until the signal dramatically drops likely due to sample precipitation (Figure 2H). These results demonstrate that optimized instrument settings significantly increase signal sensitivity and that the flexibility in sample buffer composition afforded by DI-MS affords enhances detection sensitivity through novel methods of sample solubilization.

Signal consistency and reproducibility was also estimated and the C intensity with relative standard deviation between LC-MS and DI-MS is shown in Figure 1E. Each method uses six individual replicates for a better evaluation result.

RNA contamination and LC-MS sample carryover

The absence of LC in DI-MS minimizes the possibility of sample carryover between injections in the new method. In order to directly compare sample carryover between DI-MS and LC-MS we extracted DNA from the immortalized hepatocyte cell line HepG2/C3A (10⁶ cells) and digested it into nucleosides for analysis. We note that RNA should be removed as part of the DNA extraction protocols (Figure 3A, top) in order to minimizing errors in signal extraction due to the presence of bases in the MS derived from RNA (Figure 3A, bottom). We next quantified the carryover resulting from incomplete elution of nucleosides from C18 chromatography and observed that about 2-3% of signal is retained in the blank injection that follows the sample injection (Figure 3B). Sample carryover is a common issue in liquid chromatography; some manufacturers circumvent the problem by utilizing single injection column consumables, e.g. EVOSEP with the EVOSEP One. In contrast, robots for automated direct injection that replace tips for every sample pickup, such as the Advion with the TriVersa NanoMate used in this study have no sample carryover due to the tip exchange at each injection and new nozzle utilized for each sample for electrospray ionization.³³

Figure 3 – Nucleoside carryover and stability.

(A) Extracted C and 5mC chromatograms of digested DNA extracted from HepG2 C3A (top). Extracted C and 5mC chromatograms of digested DNA extracted from HepG2 C3A treated with RNase A (bottom). (B) Extracted A, T, C, G chromatograms of digested DNA extracted from HepG2 C3A samples (top) and subsequent blank injection (bottom). (C) Nucleoside degradation test over 12 hours (D) raw chromatogram displaying the nucleosides of A, T, G, DNA C and RNA C from solubilized samples and injected once a day for 14 days (left), or (E) pre-dried nucleosides resuspended by the nLC autosampler right before the injection.

Stability issue for low throughput LC-MS analysis

Stability is an important issue for the analysis of large batches of samples that might require maintaining some samples in solution for days before they are injected into the MS. We assessed the decay rate of nucleosides in solution and demonstrate that nucleosides in solution are degrade with time and thus are no longer retainable by LC. First, we performed a stability test utilizing a mixture of synthetic unmodified nucleosides dissolved in 0.1% TFA to a final concentration of 10 μM each. The mixture of synthetic deoxynucleosides was freshly made before the test and aliquoted on 96 well plate. We injected them using LC-MS at intervals of 30 min per injection over 12 hours (Figure 3C). To estimate the abundance for each nucleoside, we summed the extracted ion chromatogram of both nucleoside and nucleobase, to avoid biases in case of variations of the nucleoside fragmentation in source. We found a relatively consistent degradation for all four nucleosides (A, T, C, G). After 12 hours, the abundance was about 60-70% compared to their initial value. Figure 3D shows the chromatograms of the decay rate of five different nucleosides: A, T, G, DNA C and RNA C observed over the course of 14 days. From the chromatograms, it is clear that the relative intensities of the nucleosides decrease with time compared to the background noise level. We repeated the experiment resuspending samples dried on a 96-well plate right before their injection (Figure 3E). These samples, although they had the same “age”, maintained a more reproducible signal intensity across the 14 days of analysis. Notably, RNA C decayed at much faster rates even if samples were dry likely due to the inherent susceptibility of RNA to hydrolysis. We also observed that the abundance of dried and then subsequently resuspended nucleoside is lower than the freshly prepared samples in the solution (Figure 3D-E, Day 0). This observation shows that drying may contribute to nucleoside decay. Together, the potential carryover by LC and the impact of time of the sample in solution are two additional factors indicating that a rapid DI-MS analysis might minimize biases in nucleoside quantification.

Conversion of observed into exact quantification ratios via calibration curves

The signal intensity of a molecule detected by LC-MS can be biased by a variety of factors, including its binding efficiency to the chromatographic column and its ionization efficiency. DI-MS acquisition can also introduce bias due to ionization efficiency of the nucleosides. To report an accurate estimation of the relative abundance of 5mC and 5hmC in samples, we generated a calibration curve to adjust the experimentally observed into actual ratio for the modified nucleosides compared to the unmodified. Varying amounts of oligonucleotides fully modified as 5mC or 5hmC were prepared using a fixed amount of oligonucleotides with only unmodified C (as mentioned in the experimental section). The samples were freshly prepared and digested for one hour before being desalted and analyzed first by LC-MS.

In LC-MS, both 5mC and 5hmC showed a linear response between the experimental and the actual ratios (Figure 4A-B). The 5mC/C ratio had a slope of 3.6096; being larger than 1 indicates that 5mC is detected with higher efficiency than C, indicating that LC-MS partially overestimates its relative abundance. This observation could be explained partially by a higher ionization efficiency of 5mC compared to C and/or by a more efficient retention of 5mC by chromatography as demonstrated by a longer retention during chromatographic separation (Figure 4C). The 5hmC/C ratio showed an inverse trend, as the curve slope of 0.3821 indicated that 5hmC is either harder to efficiently retain by LC or it has a lower ionization efficiency than C. Indeed, the 5hmC signal elutes earlier than C or mC using C18 chromatography, possibly affecting its complete retention. We have tested other types of chromatographic separations, including online PGC and ammonium formate in the buffer composition. PGC showed a very efficient retention of nucleosides, but low resolution and significant issues in carryover (Supplementary Figure S2). Salts in buffers such as ammonium formate were tested, but they rapidly accumulated on the surface of the instrument source, creating concerns on the frequency of cleaning required (data not shown).

Figure 4 – Correction of observed vs theoretical concentration of modified nucleosides.

(A) Calibration curve for 5mC% (red) and (B) 5hmC% (green) by LC-MS. (C) Extracted chromatograms of C, 5mC and 5hmC with corresponding retention time. (D) Calibration curve for 5mC% and (E) 5hmC% by DI-MS. (F) Mass spectrum with the ion intensity of C, 5mC and 5hmC. 5mC and 5hmC were magnified using Xcalibur to allow visualization due to their low abundance.

Next, we performed the same calibration curve analysis by DI-MS (Figure 4 D-E). The results highlight that both 5mC and 5hmC are quantified with an experimental ratio more similar to the actual ratio compared to the analysis performed using LC-MS. The slope of the calibration curve was closer to 1 compared to LC-MS; i.e. 2.0537 and 1.2316 for 5mC and 5hmC, respectively. The combined results of Figure 4 suggest that the low sensitivity of 5hmC calculated by LC-MS might be mostly due to the ability of retaining the compound to the column, rather than the low ionization efficiency. In fact, DI-MS quantification provides a calibration curve with a slope very close to 1. Notably, signal extraction is also computationally simpler in DI-MS acquisition, as the intensity of the signal in the mass spectrometer can be extracted just by exporting the intensity of the m/z value in a single spectrum (Figure 4F). This operation is significantly simpler than extracting the area of the extracted ion chromatogram necessary for LC-MS quantification (Figure 4C); in particular, automatic peak area integration is challenging for algorithms when the chromatogram is not distributed as a Gaussian and it is poorly discriminated from background noise.

Finally, we assessed the limit of quantification of the amount of DNA sample needed (LOQ-DNA) to accurately quantify 5mC and 5hmC if present at a fractional abundance compared to total C. To do so, we utilized an 897 bp dsDNA oligonucleotide with an approx. equal proportion of ATCG, including 5% of mC and 1% of hmC. We verified the LOQ-DNA to be 0.12 ng for 5mC and 7.5 ng for 5hmC using LC-MS (Figure 5A). This difference in sensitivity highlighted once more the difficulties in detection for 5hmC. When using DI-MS, the LOQ-DNA was assessed as 0.052 and 0.026 ng for 5mC and 5hmC, respectively. The corresponding concentration of 5% 5mC and 1% 5hmC samples are 0.04 ng/uL, 2.5 ng/uL and 0.63 ng/uL, 0.31 ng/uL for LC-MS and DI-MS respectively. The comparable LOQ-DNA for the two analytes in DI-MS demonstrates that direct injection reduces the bias in quantifying nucleosides modifications with different levels of hydrophobicity. An LOD experiment was also conducted with similar outcome; the comparison of both LOD and LOQ for individual 5mC and 5hmC with previously developed methods can be found in Supporting Information Table S1 and S2.

Figure 5 – Bias analysis of LC-MS vs DI-MS.

(A) LOQ-DNA of 5mC and 5hmC using LC-MS and DI-MS. (B) Extracted C chromatogram showing incomplete DNA digestion (dimer, GC base pair et al.). (C) Methylation level comparison between LC-MS and DI-MS. DNA was extracted from HepG2 C3A cells treated with (brown) and without (blue) RNase A.

Quantification bias due to incomplete DNA digestion

We compared quantification of the relative abundance of 5mC in HepG2/C3A cells by LC-MS and DI-MS. Digested DNA was analyzed by LC-MS to identify a potential bias in quantification when incomplete digestion occurs (Figure 5B). Specifically, incomplete DNA digestion leads to the generation of multiple LC-MS signals due to the differences in hydrophobicity of dC vs nucleoside dimers or trimers. This potentially creates a problem of reproducibility, which is not always predictable. Moreover, automated software would likely retrieve the peak area of only the main extracted ion chromatographic peak, creating issues of automation. To demonstrate that dC was present in all the highlighted peaks in Figure 5B, we show that by extracting all the peak areas we obtain a different estimation of the relative abundance of 5mC (Figure 5C). Specifically, 5mC is estimated as 2.5% with canonical signal extraction, while it is adjusted to 1.32% once all signals corresponding to C are extracted from the chromatogram in Figure 5B (top).

DI-MS quantification does not present the issue of multiple chromatographic peaks, and the presence of nucleoside dimers or trimers is circumvented by the in source fragmentation applied during the detection. The relative quantification of 5mC calculated with DI-MS (1.49%) demonstrated that LC-MS would have mistaken of almost 2-fold the estimation of the % of 5mC if not all chromatographic peaks were properly extracted. Together, issues in imperfect sample preparation harms the accuracy of LC-MS analysis more than DI-MS.

Global quantification and comparison of 5mC and 5hmC in different cell lines

To demonstrate the throughput and general accuracy of our newly developed DI-MS method, DNA was extracted from thirteen different cell lines and analyzed by both the LC-MS and DI-MS methods (Figure 6). Included in this cohort are insect, rodent, and human cell lines. The DI-MS results is highlighted by red. The LC-MS results calculated by summing all the peaks (as discussed above, Figure 5C) is highlighted by blue. These data clearly show close correspondence between the DI-MS and correctly analyzed LC-MS results. In contrast, the LC-MS results calculated by using only the major peak area (light blue), an analysis method that grossly overestimates the extent of methylation (Figure 5C), consistently yields erroneous values in each and every cell for both methylation (Figure 6A) and hydroxymethylation (Figure 6B). From the statistical analysis of these data, we conclude that DI-MS is generally as accurate as correctly analyzed LC-MS and more precise. Moreover, we aimed to demonstrate that DI-MS can robustly perform for large number of samples not only theoretically. To do so, we performed 81 injections of 13 different cell types using both DI-MS and LC-MS. The description of the samples is illustrated in Supplementary Table S3. Results showed high consistency and reproducibility for 5mC quantification in both LC-MS and DI-MS (Figure 6). The relative quantification of 5hmC showed discrepancies between the two methods (Figure 6B). As previously mentioned, the limit of detection for 5hmC is limited by the less efficient binding to C18 chromatography. For HepG2/C3A, ExpiCHO, S2 and SF9, 5hmC could only be detected by DI-MS.

Figure 6 – Relative quantification of 5mC (A) and 5hmC (B) in 13 different cell lines performed using DI-MS and LC-MS.

LC-MS quantification was performed using both the most intense extracted ion chromatographic peak and, when available, all the peaks corresponding to the same intact mass of the (un)modified nucleoside. As illustrated in Figure 5B, chromatography occasionally presents multiple signals for a given nucleoside, mostly due to incomplete DNA digestion.

The inclusion of insect cell lines within the analyzed cohort provides a control for our analysis. Published bisulfite sequencing studies³⁵ of the insect orders Diptera and Lepidoptera report methylation levels of less than 1%. Figure 6 shows that our MS analysis yields results consistent with bisulfite sequencing; minimal nucleoside modification is detected for the cell lines derived from both orders. Global levels of 5mC and 5hmC have been published³⁰ for the human kidney cell line 293T that we analyzed. The published values of 3.9% and 0.02%, respectively, closely match the values of 4.17% and 0.02% obtained by DI-MS. Another work³⁶ published recently reported global methylation level of HepG2 and HeLa to be ~1.6% and ~3.3% (1:2) comparing to our data 2.6% and 4.7% (1:1.8). They mentioned the samples were de-proteinated and desalted prior to injection into LC-MS/MS. Interestingly, when we observed RNA contamination in our HepG2 cells (Figure 5C), we calculated the methylation level to be 1.49%. It is possible that in some of the previous publications, RNase A digestion or RNA removal was not always 100% efficient.

CONCLUSIONS

We report the optimization and assessment of a DI-MS platform that performs rapid and unbiased relative quantification of global levels of DNA cytidine methylation (5mC) and hydroxymethylation (5hmC). This approach reduces instrument analysis time to less than 1 min per sample. As a result, this method can potentially analyze the global levels of nucleoside modifications for thousands of DNA samples per day. DI-MS obviates the confounding issues of LC-MS including biased analyte retention, sample carryover and limited choices in the use of solubilization buffers that provide efficient detection. DI-MS utilizes very small amounts of solutions (2-10 μL) per sample and generates very little solvent waste.

DI-MS is simpler to conduct compared to LC-MS as it requires no chromatographic optimization and the analyte intensity is determined from the intensity of the spectrum rather than from extracting ion chromatograms. Although RNA removal is important to both methods, it is a critical component of the DI-MS sample preparation protocol. As the DI-MS acquisition protocol fragments nucleosides into nucleobases; the presence of RNA would overestimate the abundance of DNA nucleoside. While the sensitivity to 5mC of LC-MS is greater than DI-MS, the sensitivity of DI-MS is independent of nucleoside type and thus provides an unbiased analysis platform for the analysis of nucleobase modification.

The DI-MS method can greatly enable basic as well as clinical research due to the possibility of analyzing large sample sets with great accuracy. Epigenomics is a fast-growing field and has made many strides in recent past. For most of the diseases, even those with a strong genetic component, there are significant upstream or downstream epigenetic changes. Whether these epigenetic changes are the cause or consequences of genetic changes is currently subject of intense active investigations. Clearly, epigenetics plays significantly important role in modifying the disease and thus its etiology. Quantitation of the amount of epigenetic marks can support resource intensive approaches that localize modification sites, such epigenome-wide association study (EWAS), to identify the new risk factors that might otherwise be missed by conventional genetic epidemiological approaches. Large-scale analysis of DNA methylation and other potential nucleoside modifications including RNA modification^{37, 38} could pave the way to the rapid and confident identification of biomarkers for diseases affecting chromatin state.

• We present an ultra-rapid (<1 min) mass spectrometry method to analyze major DNA modifications.

• The method overcomes typical issues of liquid chromatography such as carryover and biased retention of nucleosides that have different chemical properties, e.g. different hydrophobicity.

• We demonstrate that routinely used targeted approaches overlook the formation of nucleoside byproducts, which affect the accuracy of the quantification.

• We provide optimized instrument settings to minimize biases in the quantification of methyl cytidine and hydroxymethyl cytidine.