Single Neuron RNA Modification Analysis by Mass Spectrometry: Characterizing RNA Modification Patterns and Dynamics with Single-Cell Resolution

Abstract

The entire collection of post-transcriptional modifications to RNA, known as the epitranscriptome, has been increasingly recognized as a critical regulatory layer in the cellular translation machinery. However, contemporary methods for the analysis of RNA modifications are limited to the detection of highly abundant modifications in bulk tissue samples, potentially obscuring unique epitranscriptomes of individual cells with population averages. We developed an approach, Single Neuron RNA Modification Analysis by Mass Spectrometry (SNRMA-MS), that enables the detection and quantification of numerous post-transcriptionally modified nucleosides in single cells. When compared to a conventional RNA extraction approach that does not allow detection of RNA modifications in single cells, SNRMA-MS leverages an optimized sample preparation approach to detect up to 16 RNA modifications in individual neurons from the central nervous system of Aplysia californica. SNRMA-MS revealed that the RNA modification profiles of identified A. californica neurons with different physiological functions were mostly cell specific. However, functionally homologous neurons tended to demonstrate similar modification patterns. Stable isotope labeling with CD₃-Met showed significant differences in RNA methylation rates that were dependent on the identity of the modification and the cell, with metacerebral cells (MCCs) displaying the fastest incorporation of CD₃ groups into endogenous RNAs. Quantitative SNRMA-MS showed higher intracellular concentrations for 2’-O-methyladenosine and 2’-O-methylcytidine in homologous R2/LPl1 cell pairs than in MCCs. Overall, SNRMA-MS is the first analytical approach capable of simultaneously quantifying numerous RNA modifications in single neurons and revealing cell-specific modification profiles.

Graphical Abstract

INTRODUCTION

Post-transcriptional modifications are naturally occurring transformations to the primary structure of RNA that range in complexity from methylation to conjugation with macromolecules.^1–3 Approximately 150 structurally unique RNA modifications have been discovered to date that are deposited reversibly^4,5 and substoichiometrically^6,7 on coding and non-coding RNAs alike. The landscape of modified nucleosides in any given organism tunes the stabilities,^8,9 folds,¹⁰ and translation activities^11,12 of endogenous RNAs, collectively forming what is known as the epitranscriptome.

Epitranscriptomic marks are unequally distributed across tissue types,^13,14 with region-specific RNA modification profiles particularly evident within the central nervous system (CNS).¹⁵ For example, defined brain regions in vertebrate models display time-dependent changes in N6-methyladenosine (m⁶A) levels during CNS development,¹⁶ exposure to acute stress,¹⁷ and memory consolidation.^18–20 Although these studies highlighted a relationship between m⁶A abundance and CNS plasticity, they were limited to the investigation of a single, highly abundant RNA modification in relatively large brain regions. However, the CNS is not only divided into distinct subregions, but further encompasses a heterogeneous consortium of individual cells that exhibit characteristic and often dynamic molecular compositions.^21–23 Such differences may emerge from local microenvironments,^24,25 molecular organizations,²⁶ or stochastic influences,²⁷ rendering population-averaging measurements insufficient for capturing the chemical diversity of single cells.^28,29 While the analysis of select RNA modifications in bulk tissue samples has become routine, bulk measurements only generate an ensemble average of the abundances and dynamics of modified nucleosides, potentially concealing the unique RNA modification profiles of morphologically and functionally similar cells in the CNS. Therefore, the field of epitranscriptomics would benefit from new approaches that facilitate the detection of numerous modified nucleosides in single neurons and elucidate cell-specific distribution of RNA modifications.

Liquid chromatography (LC)–mass spectrometry (MS) approaches are ideally suited for simultaneously providing quantitative information on RNA modifications for the entire collection of modified nucleosides in an organism.^30–32 We recently developed an LC–MS-based method optimized for CNS tissues in the neurobiological model animal, Aplysia californica.³³ Using this approach, we found that a non-associative learning paradigm in A. californica produced time-dependent shifts in neuronal RNA modification profiles across CNS subregions (i.e., major ganglia).³⁴ These characteristic changes in neuronal RNA modifications contributed to increased synthesis of poly-glutamine proteins and neuron excitability, reinforcing the functional role of multiple RNA modifications in CNS plasticity. However, the investigation was limited to profiling modified RNAs in whole ganglia (~10³ cells) and likely obscured the RNA modification dynamics of individual neurons within the activated neural circuits. While the judicious selection of RNA sample preparation and analysis conditions can be undertaken to maximize RNA yields from small-volume samples,^35,36 characterizing the full complement of RNA modifications in single cells remains a formidable challenge.

Here, we introduce an approach, Single Neuron RNA Modification Analysis by Mass Spectrometry (SNRMA-MS), that simultaneously detects and quantifies dozens of RNA modifications in individual cells. SNRMA-MS leverages a trituration and heat-based sample preparation protocol to liberate sufficient RNA from single, identified neurons in the CNS of A. californica for downstream LC–MS analysis. By comparison, conventional RNA extraction using phenol-chloroform yielded no detectable modified nucleosides from single neurons. Qualitative and quantitative SNRMA-MS experiments revealed that the post-transcriptional RNA modification profiles of identified neurons with different physiological functions were distinguishable from each other, while functionally homologous neurons tended to have similar evaluated epitranscriptomic modification patterns. We combined SNRMA-MS with stable isotope labeling of nucleic acids to uncover dramatic differences in RNA methylation rates that were dependent on the identity of both the modification and the cell. Taken together, these results provide the first evidence of cell-specific distribution of post-transcriptionally modified nucleosides and establish SNRMA-MS as a tool capable of delivering deeper insight toward understanding the mechanisms and functional implications of RNA modification heterogeneity in individual neurons in the CNS.

EXPERIMENTAL

Important experimental details are provided here, with further experimental details found in the Supporting Information.

Reagents.

All reagents were used without further purification (see Supporting Information).

Animals and Single Neuron Isolation.

A. californica (150–250 g) were purchased from the National Resource for Aplysia (Miami, FL) and maintained in an aquarium with circulating artificial seawater (ASW, Instant Ocean, Blacksburg, VA) at 14 °C. Animals were anesthetized by injection of a 0.33 M MgCl2 solution (14 °C) into the body cavity with a solution volume (in mL)-to-animal body mass (in g) ratio of 1:3. Major CNS ganglia were surgically isolated and treated with protease type XIV from Streptomyces griseus (Sigma-Aldrich, St. Louis, MO) at 34 °C for 30 min or 1 h (for cerebral ganglion). The ganglia were rinsed six times with ASW and pinned in a Sylgard-coated dish. The ganglion sheaths were then carefully removed to expose neuron cell bodies. Sharp glass capillaries or sharp tungsten needles were used to isolate identified neurons from the bulk ganglia. The isolated cells were then aspirated with a plastic micropipette and transferred under visual control into a PCR tube containing 4 μL of 0.365 M ammonium acetate. Aliquots of ASW were collected from the Sylgard-coated dish to serve as blank measurements.

In situ Culture and Stable Isotope Labeling.

Ganglia were isolated from anesthetized animals, rinsed with ASW, placed in 12-well plates containing 2.5 mL of ASW with antibiotics (see Supporting Information), and incubated at 12 °C for 2–6 d. For stable isotope labeling experiments, CD₃-Met was added to a final concentration of 0.5 mM. The CD₃-containing culture medium was replaced every 2 d. For RNA synthesis inhibition, 2.5 μL actinomycin D stock (50 mg mL⁻¹ in dimethyl sulfoxide) was added to the culture medium. Neurons were then isolated (vide supra) and subjected to SNRMA-MS.

SNRMA-MS: Cell Lysis, RNA Digestion, and LC–MS/MS.

Isolated neurons were triturated by repeated aspiration and dispensing with a micropipette (~100 μm inner diameter) in 0.365 M ammonium acetate at room temperature. For cells that didn’t immediately rupture, a pulled glass capillary (~10 μm outer diameter) was used to manually lyse the neuron by applying pressure across the diameter of the cell. Lysed cells were then heated at 95 °C for 3 min to facilitate disruption of the cell membrane and deactivate endogenous enzymes. After cooling the solution to 10 °C, liberated RNA was digested with minimized sample volume: 1 μL of 10 μg μL⁻¹ bovine serum albumin, 0.5 μL of 0.5 μg μL⁻¹ pentostatin, 0.495 μL of 2 U μL⁻¹ alkaline phosphatase, 1 μL of 0.1 U phosphodiesterase I (in 10 mM MgCl2), and 0.38 μL benzonase (25 U). RNA was then hydrolyzed at 37 °C for 3 h and approximately 7 μL of the solution was carefully transferred into an autosampler vial for instrumental analysis.

RNA digests from single neurons and whole ganglia were analyzed by a Dionex Ultimate 3000 nanoLC system (Thermo Scientific, Waltham, MA) and an Impact HD UHR QqTOF mass spectrometer (Bruker Corp., Billerica, MA). For the separation of nucleosides, 4 μL of sample was injected onto an Acclaim RSLC 120 C18 column with the dimensions 150 × 2.1 mm, 2.2 μm (Thermo Scientific) and separated at a flow rate of 0.2 mL min⁻¹ and 36 °C. Gradient elution and MS were performed as previously described.^33,34

Data Analysis.

Modified nucleosides were identified by comparing MS2 spectra and LC retention characteristics to database values.¹ Peak areas were obtained by manual integration of corresponding regions of extracted ion chromatograms (EICs) for modified nucleosides (m/z from MODOMICS¹) using Compass DataAnalysis 4.4 software (Bruker). Consistent with signal normalization strategies previously reported in single cell analysis,^37–39 peak areas were normalized to endogenous nucleosides by dividing by the sum of peak areas for canonical cytidine, uridine, and guanosine in the sample. Adenosine was not included in the normalization because of its role in the CNS as a dynamic neuromodulator⁴⁰ and also because its signal saturated the detector in the abdominal ganglia extracts. Principal component analysis (PCA) was performed in RStudio⁴¹ for normalized peak areas of modified nucleosides with signal to noise ratios >10. OriginPro 2021b software (OriginLab Corp., Northampton, MA) was used for pairwise comparisons and for constructing linear calibration curves.

RESULTS AND DISCUSSION

Simultaneous Detection of Numerous RNA Modifications in Single Neurons by SNRMA-MS.

A. californica is an experimentally advantageous model for neurobiology because of its relatively large neurons (10–1000 μm diameter), which are easily identifiable from one animal to the next, and exhibit well-defined functions, morphology, and biochemistry.⁴² Capitalizing on these features of the A. californica CNS, we developed SNRMA-MS, wherein identified neurons are manually isolated and subjected to a small-volume (~5 μL) sample preparation approach that renders numerous modified nucleosides detectable by LC–MS/MS (Figure 1A–E). From single identified neurons known as LPl1 (diameter ~500 μm), SNRMA-MS detected 15 ± 1 modified nucleosides that were confirmed by exact masses, LC retention times, and/or MS2 fragmentation patterns (n = 3, Figure 1D and Figure S1A–P). In order to test whether the observed modified nucleosides were derived from endogenous RNA biopolymers, extracellular RNA, or free intra/extracellular modified nucleosides, two control experiments were performed: 1) an aliquot of the medium surrounding the ganglia (ASW with antibiotics) was subjected to digestion and LC–MS analysis and 2) an isolated LPl1 neuron was lysed and incubated without enzymes prior to analysis by LC–MS. Canonical nucleosides, but not modified nucleosides, were detected in both controls (Figure S2 and Table S2), indicating that the modified nucleosides detected by SNRMA-MS originated from intracellular RNAs.

Figure 1.

Simultaneous detection of numerous RNA modifications in single neurons using SNRMA-MS. (A) Desheathed buccal hemiganglion and identified B1 cell (white arrow). (B) Isolated B1 cell. Scale bar = 220 μm. (C) Isolated neuron in sample tube prior to digestion. (D) EICs for RNA modifications detected in a single LPl1 neuron (see Table S1 for m/z values used for EICs and modified nucleoside abbreviations). (E) SNRMA-MS workflow. (F) Comparison of EIC peak areas for RNA modifications in single LPl1 neurons using different sample preparation approaches. Error bars indicate standard deviations, n = 3, nd = not detected. Unpaired t-test with Bonferroni-Holm correction, *p < 0.05.

We then optimized the SNRMA-MS approach and compared it to the results of a conventional phenol-chloroform-based single-neuron RNA extraction protocol.^43,44 Initially, cell lysis was performed by trituration at room temperature, followed by immediate digestion of RNA at 37 °C. This approach resulted in significantly lower signals for numerous RNA modifications compared to when we included a heating step before RNA digestion (Figure 1F) to improve digestion efficiency, perhaps due to destabilization of RNA secondary structures.⁴⁵ We therefore elected to heat the sample at 95 °C for 3 min in further experiments. To ensure that modified nucleosides were not artificially generated due to sample preparation, we spiked 3 mM stable isotope-labeled methionine (CD₃-Met, a cofactor for RNA methyltransferases) into single-neuron samples and performed SNRMA-MS. No CD₃-labeled nucleosides were detected (Figure S3), demonstrating that nucleoside modifications were not induced by the SNRMA-MS workflow.

Compared to a conventional phenol-chloroform RNA extraction procedure that produced no detectable RNA modifications, SNRMA-MS detected 15 ± 1 modified nucleosides (Figure 1F). These results show that over half of the known set of epitranscriptomic modifications in the A. californica CNS³⁴ can be reliably detected in an individual neuron using SNRMA-MS. With advances in sampling and separation methods that facilitate handling of small volumes,^35,46–48 we expect that near complete epitranscriptome coverage for smaller cells will be achievable.

Comparison of RNA Modification Profiles for Single Neurons and Bulk Tissue.

Next, we leveraged SNRMA-MS to investigate whether whole-ganglion measurements of post-transcriptional modifications are representative of individual neurons within the same ganglion. We compared the relative abundances of RNA modifications in the abdominal ganglion to an identified cholinergic neuron, R2, isolated from the same ganglion. Identical sample preparation methods were used for isolated R2 neurons and the remaining abdominal ganglia to ensure that no bias was introduced due to differences in analytical workflows (Figure 2A). In total, 19 ± 1 modified nucleosides were detected in the abdominal ganglia (n = 7, Figure S4), 12 ± 2 of which were also detected in single R2 neurons (n = 7, Table S3). The relative standard deviation (RSD) values for measured abundances of the RNA modifications detected in both R2 neurons and bulk abdominal ganglia ranged from 14% to 72% and 7% to 70%, respectively. Overall, mean RSD values for the detected RNA modifications over n = 7 biological replicates were 33 ± 20% and 21 ± 17% for single neurons and ganglia, respectively. These data demonstrate that SNRMA-MS exhibits excellent precision that is comparable to bulk cell measurements of modified nucleosides.

Figure 2.

Distinguishing single R2 neurons identified from bulk CNS tissue on the basis of modified nucleoside abundances. (A) Abdominal ganglion and R2 neuron (Fast Green dye-injected) in the A. californica CNS were subjected to SNRMA-MS. (B) PCA score (top) and loading plots for normalized peak areas of 13 RNA modifications. Each number on the score plot refers to a different animal from which R2 neurons and abdominal ganglia were derived. (C) Comparison of 13 RNA modifications in single R2 neurons and surrounding ganglia from a second cohort of animals (n = 7). Error bars are standard deviations, *p < 0.05, ***p < 5E–4, paired t-test with Bonferroni-Holm correction for multiple comparisons. (D) Subset of RNA modifications from panel C showing R2–abdominal ganglion pairs for each animal.

Principal component analysis (PCA) of normalized peak areas for 13 RNA modifications showed an obvious distinction between R2 neurons and their ganglia counterparts (Figure 2B). From the loading plot, the cell-ganglion pairs were primarily separated due to differential abundances of the methyladenosine positional isomers, pseudouridine (Ψ), and 2’-O-methylguanosine (Gm). To validate the PCA results, R2 neurons and abdominal ganglia from a second cohort of animals (n = 7) were analyzed by SNRMA-MS and the detected modifications subjected to pairwise comparisons. Significantly lower relative abundances of Ψ and Gm were observed in R2 neurons compared to the abdominal ganglion, whereas levels for 2’-O-methyladenosine (Am) trended higher in R2 neurons but were not significantly different (Figure 2C). Further inspection showed that all R2 neurons exhibited lower abundances of Ψ and Gm compared to their corresponding ganglia and all but one of the cell-ganglion pairs showed higher levels of Am in R2 neurons (Figure 2D). Taken together, the PCA results and pairwise comparisons demonstrate that the R2 neuron deviates from the bulk abdominal ganglion on the basis of post-transcriptionally modified nucleosides. While it is known that individual cells can exhibit molecular phenotypes distinct from the population average, these are the first results to demonstrate that RNA modifications can be heterogeneously distributed across cells in the same tissue.

RNA Modification Signatures of Functionally Different Neurons.

We then asked whether functionally distinct neurons exhibited unique RNA modification chemistries. For these experiments, we compared the modified nucleoside profiles of four identified cells: the metacerebral cells (MCCs, two cells per animal), B2 neurons (two cells per animal), LPl1 (single cell per animal), and R2 (single cell per animal). These neurons were either freshly isolated or, when testing the impact of cell culture on modification profiles, cultured in their respective ganglia (i.e., in situ) for 48 h prior to isolation and subsequently subjected to SNRMA-MS. Signals for a total of six RNA modifications were above the lower limit of quantification (LLOQ) threshold (signal-to-noise > 10) in all of the studied neurons, were normalized to canonical G, C, and U abundances, and used as inputs for PCA. The PCA score plot shows distinct separation of individual neurons based on their functional identities (Figure 3). The MCCs, which are serotonergic neurons in the cerebral ganglion that regulate feeding behavior,⁴⁹ exhibited unique RNA modification profiles as evidenced by their separation from B2 neurons and R2/LPl1 neurons. Similarly, B2 neurons, which are peptidergic cells in the buccal ganglion involved in gut motility,⁵⁰ formed an obvious cluster in the score plot. The functionally homologous R2 and LPl1 neurons, which both control mucus release,⁵¹ displayed similar modified nucleoside contents and therefore co-clustered.

Figure 3.

RNA modification profiles of identified neurons. (A) PCA score and (B) loading plots derived from normalized peak areas for six RNA modifications detected in MCCs, and B2, R2, and LPl1 neurons. Freshly isolated neurons were compared to neurons isolated from 48 h in situ cultures, which are denoted by cMCC, cB2, cR2, and cLPl1. One of the two cLPl1 neurons directly overlapped with a cR2 neuron in the score plot.

SNRMA-MS revealed that PCA cluster formation based on neuron function (e.g., homologous R2 and LPl1 neurons) was maintained even after in situ culture for 48 h, albeit with slight differences in RNA modification statuses compared to freshly isolated cells. These results are similar to differences in the metabolomes of cultured and freshly isolated A. californica neurons.⁵² The modified nucleoside profiles of cultured neurons were also more similar than freshly isolated neurons, evidenced by smaller cluster radii in the PCA score plot. This can likely be explained by the native neuron microenvironment that promotes cell-cell heterogeneity compared to neurons in culture,²⁵ perhaps via neuronal network regulation by numerous synaptic inputs. Overall, SNRMA-MS showed that single neurons are distinguishable by their RNA modification compositions, which appear to be dependent on neuron identity and function.

Investigating RNA Modification Dynamics in Single Neurons with Stable Isotope Labeling.

Although tremendous progress in understanding RNA modification incorporation rates has been made using bulk cell culture experiments, the dynamics of RNA methylation in individual cells have yet to be reported. To study RNA methylation events with single cell resolution, we first isolated pleural ganglia from the A. californica CNS and cultured them in ASW in the presence of CD₃-Met for 2–6 d at 12 °C (Figure S5A). After the 6-d culture, SNRMA-MS detected 7 ± 2 CD3-methylated nucleosides in LPl1 neurons, with five CD₃-methylated nucleosides detected in all neurons studied (Figures S5B and S6). The peak area ratios of CD₃-labeled nucleosides to their corresponding unlabeled methylated nucleosides were highest for N1-methyladenosine (m¹A) and 5-methylcytidine (m⁵C), with ~50% of the detectable nucleosides bearing a CD3 group. We then compared the dynamics of adenosine methylation over the course of 2–6 d. No discernible difference in CD₃ incorporation was observed in the time course until day 6, where a significantly higher CD₃ labeling ratio was observed for m⁶A compared to Am (Figure 4A). Significantly higher CD3 labeling was also observed for m¹A compared to m⁶A (p < 0.05, unpaired t-test) and Am (p < 0.05, unpaired t-test) (Figure S6). These data indicate that the methylation rate is position-specific in the cultured LPl1 neurons. Since the signals for CD3-labeled and unlabeled nucleosides originated from total RNA, the subtype(s) of RNA (i.e., rRNA, tRNA, mRNA) that contribute to differential incorporation of CD₃ for these positional isomers cannot be distinguished. However, because m¹A and m⁶A are more commonly found in mRNA and tRNA, and Am is predominantly incorporated in rRNA, the slower turnover of rRNA relative to mRNA and tRNA⁵³ may be responsible for the observed differences in CD₃ labeling rates.

Figure 4.

RNA modification dynamics in single neurons by stable isotope labeling. (A) Peak area ratios for CD₃-labeled/unlabeled m⁶A and Am in LPl1 neurons over a 6-d time course. (B) Peak area ratios for CD₃-m¹A/unlabeled m¹A over a 6-d culture for MCCs, and R2 and LPl1 neurons. Error bars are standard deviations, n = 3 for each data point with 30 total neurons examined for (A) and 45 neurons for (B), * p < 0.05, ** p < 0.005, unpaired t-test.

Another explanation for the differential labeling of the positional isomers of methyladenosine is that the m¹A and m⁶A modifications are known to be reversible in tRNA and mRNA,^4,54 which may facilitate the incorporation of CD₃ into mature RNAs that undergo demethylation. To test this hypothesis, we treated pleural ganglia with both CD3-Met and the RNA synthesis inhibitor actinomycin D,⁵⁵ and evaluated CD₃ labeling in the LPl1 neurons by SNRMA-MS. After the 6-d culture with actinomycin D, we did not observe any CD₃ labeling of m¹A, m⁶A, or any other methylated nucleosides in LPl1 (Figure S7). These results demonstrate that faster labeling of m¹A and m⁶A compared to Am is likely from RNA methylation events occurring during RNA synthesis, rather than from mature RNAs in the cytosol.

Another advantage of coupling SNRMA-MS with stable isotope labeling is that it allows investigation of cell-specific RNA modification dynamics. Since we observed the fastest CD₃ incorporation rates for m¹A, we studied CD₃-m¹A labeling in R2 and LPl1 neurons, which are nearly identical in terms of function, electrophysiology, and pharmacological properties; however, their mRNA expression profiles delineate as the animal ages.⁵⁶ We also compared these cells to the functionally distinct MCCs over a time course of 6 d. While no differences in labeling rates were observed initially, the three neurons displayed significantly different CD3-m¹A/m¹A ratios at the 4-d time point, with MCCs exhibiting the fastest incorporation rate and R2 the slowest (Figure 4B). No obvious differences in intracellular CD₃-Met levels were observed for the three identified neurons (Figure S8). These results emphasize the utility of combining SNRMA-MS and stable isotope labeling to uncover distinctive neuronal RNA modification chemistries.

Quantification of Modified Nucleosides in Single Neurons.

Quantitative SNRMA-MS was then performed for five modified nucleosides in single neurons. We used external calibration curves generated with authentic standards, similar to previous efforts for quantifying metabolites in single cells.^37,57 Linear calibration curves (R² > 0.99) for m¹A, m⁶A, Am, Ψ, and 2’-O-methylcytidine (Cm) were constructed over two orders of magnitude with LLOQ values ranging from 0.1 pmol for m¹A to 1 pmol for Cm. In A. californica, the cerebral ganglion can be divided into two approximately symmetrical left and right hemiganglia, each containing a single MCC (Figure 5A, LMCC and RMCC, respectively) that is biochemically, morphologically, and functionally homologous to its counterpart. R2 and LPl1 neurons are also functionally homologous but reside in different ganglia (Figure 5B). This feature of the A. californica CNS allowed us to ask whether these symmetrical neurons display similar RNA modification quantities.

Figure 5.

Quantification of RNA modifications in single neurons. Photographs of (A) cerebral ganglion showing left and right MCCs (circled). (B) Photographs of R2 and LPl1 neurons in desheathed regions of the abdominal and pleural ganglia, respectively. Quantification of (C) m¹A and (D) Ψ in symmetrical MCCs and R2/LPl1 neurons. Pairs of cells are derived from the same animal and triangles indicate calibration standards. (E) Comparison of intracellular concentrations for select RNA modifications in MCC and R2/LPl1 cell pairs. Data points are connected by thick lines for better visibility. * p < 0.05, ** p < 0.005, paired t-test with Bonferroni-Holm correction for multiple comparisons, n = 2 animals for MCCs (4 cells total) and n = 3 animals for R2/LPl1 (6 cells total).

Using SNRMA-MS, we measured the quantities of five modified nucleosides in freshly isolated pairs of MCCs and R2/LPl1 neurons from two and three animals, respectively. As expected, the intracellular quantities for all modified nucleosides (except Cm) in symmetrical MCC pairs were nearly identical (Figures 5C, D and S9A–C), with a minimum observed difference in RNA modification abundance of 2.1% for Am and a maximum difference of 13.1% for Ψ between paired cells. In contrast, R2 and LPl1 neuron pairs showed higher and more variable total quantities of the studied modified nucleosides. For example, the LPl1/R2 pair from animal 1 displayed 0.29 and 0.50 pmol of m¹A, corresponding to a 41.4% difference in quantity of this modification between these functionally homologous neurons.

We then asked if the observed differences in total modified nucleoside quantities between LPl1 and R2 neurons were due to variability in cell volume and would thus be corrected if intracellular concentrations of modified nucleosides were compared. Optical analysis of neurons subjected to SNRMA-MS showed that while MCC pairs were generally similar in diameter, R2 and LPl1 were less congruent in their cellular dimensions (Table S4), so we used ellipsoid estimations of cell volume. Calculated intracellular concentrations of m¹A, m⁶A, Am, Ψ, and Cm were between ~2–20 μM for MCCs and R2/LPl1 neurons, with Am and Cm significantly higher in R2/LPl1 neurons compared to MCCs (Figure 5E). Higher variability was observed for R2/LPl1 cell pairs, perhaps due to errors in measuring cell diameter (propagated cubically) or real biological differences between R2 and LPl1. These results from SNRMA-MS represent the first quantitative analysis of multiple RNA modifications in single cells and highlight the importance of considering functionally similar cells as potentially heterogeneous populations.

CONCLUSIONS

We have developed the SNRMA-MS approach that, for the first time, simultaneously quantifies numerous modified ribonucleosides within single cells. Our method revealed cell-specific RNA modification profiles for single neurons that are distinct from bulk surrounding tissue. Coupling SNRMA-MS with stable isotope labeling permitted the comparison of RNA methylation dynamics, which were determined to be both modification- and cell-dependent. Quantitative SNRMA-MS provided further insight toward the unique RNA modification chemistries of neurons with single-cell resolution.

The ability to characterize epitranscriptomic marks with single-cell resolution is expected to provide unprecedented information about cell-specific post-transcriptional regulation in otherwise homogeneous cell populations. One can envision that similarly heterogeneous RNA modification profiles exist in mammalian cells, which, due to their smaller size, will require improved sampling and separation approaches. One limitation of the current approach involves the required steps for extracting and handling the samples, currently limited to about 5 μL; we expect that by incorporating microfluidic manipulations, one could integrate the various steps into a dedicated device that requires much smaller volumes. These advancements will also require improved sample preparation approaches that mitigate endogenous nuclease activities and enhance the relatively lower signals observed for RNA fragments in MS detection.⁵⁸ With such a system, we expect that the minimum cell size can be decreased at least two- to three-fold, enabling the approach to be used with larger mammalian neurons, e.g., the dorsal root ganglion cells, Purkinje neurons, and magnocellular neurons. When integrated with nanoscale separation methods,^35,46 even smaller cells will become assessable.