Bisulfite and Nanopore Sequencing for Pseudouridine in RNA

Conspectus:

Nucleophilic addition of bisulfite to pyrimidine bases has been known for a half century, and the reaction has been in use for at least a quarter of a century for identifying 5-methylcytidine in DNA. This account focuses on the chemistry of bisulfite with pseudouridine, an isomer of the RNA nucleoside uridine in which the uracil base is connected to C1’ of ribose via C5 instead of N1. Pseudouridine, Ψ, is the most common nucleotide modification found in cellular RNA overall, in part due to its abundance in ribosomal RNAs and transfer RNAs. It has a stabilizing influence on RNA structure because N1 is now available for additional hydrogen bonding and because the heterocycle is slightly better at π stacking. The isomerization of U to Ψ in RNA strands is catalyzed by 13 different enzymes in humans and 11 in E. coli; some of these enzymes are implicated in disease states which is testament to the biological importance of pseudouridine in cells. Recently, pseudouridine came into the limelight as the key modification that, after N1 methylation, enables mRNA vaccines to be delivered efficiently into human tissue with minimal generation of a deleterious immunogenic response. Here we describe the bisulfite reaction with pseudouridine which gives rise to a chemical sequencing method to map the modified base in the epitranscriptome. Unlike the reaction with cytidine, the addition of bisulfite to Ψ leads irreversibly to form an adduct that is bypassed during cDNA synthesis by reverse transcriptases yielding a characteristic deletion signature. Although there were hints to the structure of the bisulfite adduct(s) 30 and 50 years ago, it took modern spectroscopic and computational methods to solve the mystery. Raman spectroscopy along with extensive NMR, ECD and computational work led to the assignment of the major product as the (R) diastereomer of an oxygen adduct at C1’ of a ring-opened pseudouridine. Mechanistically, this arose from a succession of conjugate addition, E2 elimination, and a [2,3] sigmatropic rearrangement, all of which are stereo-defined reactions. A minor reaction with excess bisulfite led to the (S) isomer of a S-adducted SO₃⁻ group. Understanding structure and mechanism aided the design of a Ψ-specific sequencing reaction and guided attempts to improve the utility and specificity of the method. Separately, we have been investigating the use of nanopore direct RNA sequencing, a single-molecule method that directly analyzes RNA strands isolated from cells after end-ligation of adaptor sequences. By combining the electrical current and base-calling data from the nanopore with dwell-time analysis from the helicase employed to deliver RNA to the nanopore, we were able to map Ψ sites in nearly all sequence contexts. This analysis was employed to find Ψ residues in the SARS-CoV-2 vRNA, to analyze the sequence context effects of mRNA vaccine synthesis via in vitro transcription, and to evaluate the impact of stress on chemical modifications in the E. coli ribosome. Most recently, we found that bisulfite treatment of RNA leading to Ψ adducts could modulate the nanopore signal to help in mapping modifications of low occupancy.

Graphical Abstract

1. INTRODUCTION

A student’s question after class in 2007 turned into a decades-long research project in our laboratory when Vahid Khoddami, then a PhD student in Oncological Sciences with Brad Cairns, came to one of us (CJB) asking for help with adapting the bisulfite sequencing protocol for locating 5-methylcytosine in DNA to sequencing for this modified base (m⁵C) in RNA. The reaction involves heating the DNA or RNA sample with 3 M bisulfite at low pH to facilitate conjugate addition of HSO₃⁻ to C6 of cytidine followed by heating at higher pH to complete deamination and subsequent bisulfite elimination, regenerating the pyrimidine base in which C has been converted to U (Scheme 1a). Sequencing the strands before and after bisulfite treatment reveals the presence of a methylated C because this modified base is 50-fold less reactive with bisulfite and does not undergo conversion to U; rather, it continues to amplify as a C.

Scheme 1.

The pH 5 bisulfite reaction on RNA induces (a) deamination of C to U, (b) reversibly reacts with U, and (c) reacts with Ψ to yield two ribose adducts.

Because of the sensitivity of RNA to heat and extremes in pH, the bisulfite reaction conditions needed to be adjusted for RNA compared to the well-known procedure used for DNA. We did that. Vahid then set about applying the protocol to map modification sites in the HeLa transcriptome. In the course of his analysis, a very similar protocol was published by others,⁵ but Vahid continued to explore. Notably, he returned to us after some time with a new question: Why and how does bisulfite react with pseudouridine? He had discovered that the RNA sequencing method for m⁵C with bisulfite not only resulted in C deamination to U, but in the same experiment, pseudouridines (Ψs) in the RNA strand were sequenced as deletions after reverse transcription and PCR amplification. What chemistry of bisulfite was responsible for this?

This account relates our unraveling of the mysterious bisulfite reaction with pseudouridine in comparison to other pyrimidines, and its use in chemical sequencing of Ψ in RNA. Next, we describe our parallel efforts at using nanopore technology to sequence RNA for Ψ, and finally, very recent efforts are described that marry the two sequencing approaches by using bisulfite modification of Ψ to enhance and optimize the signals from nanopore direct RNA sequencing.

2. THE BISULFITE REACTION WITH PSEUDOURIDINE

2.1. Chemical Structures of Bisulfite Adducts

The role of aqueous bisulfite in the reaction with cytosine is to add reversibly to C6, destroying the aromaticity of the pyrimidine base which then leads to hydrolytic deamination of the N⁴ amino group.^6,7 The addition reaction is facile at pH 5 at elevated temperatures, and later the desulfonation to re-aromatize after deamination proceeds at pH 9. Uracil, the heterocyclic base of both uridine and pseudouridine, cannot deaminate, but it clearly undergoes a reaction that leads to a major change when an RNA strand is reverse transcribed. However, only pseudouridine and not uridine leads to the deletion signature, hinting that the connectivity between ribose and the base must be responsible for the change in chemical outcome of the bisulfite reaction (Scheme 1b,c).

Two early reports in the literature gave us part of the answer. First, work by Singhal in 1974 showed that addition of bisulfite to Ψ led to two stable products that both retained the UV absorbance characteristic of pseudouridine’s aromaticity (λ_max = 260 nm).⁸ It was proposed that the two adducts were base adducts, but no specifics about the structures were given. A second hypothesis for the structure of the bisulfite adducts was proposed in the 1980 PhD dissertation of Donald Everett, a student of Robert Shapiro at NYU.⁹ Again, two stable adducts were isolated and characterized, and the structures proposed now involved C1’ connectivity to bisulfite, presumably a pair of diastereomers (Fig. 1). Troubling points in the structure elucidation included an accidental coupling constant of zero in the ¹H NMR spectrum for J_H1’H2’ for one of the compounds, and no information was available about the connectivity of bisulfite to C1’, S vs. O, or about absolute configurations at C1’.

Figure 1.

The HPLC chromatogram for the pH 5 bisulfite reaction with Ψ nucleoside monitored at 220 nm (red) and 260 nm (blue).

We built on these foundational studies by following the reaction course of pseudouridine ribonucleoside with bisulfite as monitored by HPLC (Figure 1).² Initially, six new peaks were observed with UV monitoring at 220 nm, all of which had masses corresponding to Ψ + HSO₃⁻. Peaks 1 and 5, both with λ_max = 260 nm, were stable adducts that grew in amount after heating. Peaks 2, 3, 4, and 6 were only observable at 220 nm at early reaction times, and upon heating, peaks 2 and 6 diminished as these compounds were converted back to Ψ. On the other hand, peaks 3 and 4 also diminished on heating, and they converted to a mixture of stable products 1 and 5.

Because compounds 1 and 5 had quite different retention times, we suspected they were not diastereomers as previously thought but rather constitutional isomers. We next sought to establish the connectively of the bisulfite moiety to C1’ for both 1 and 5 by conducting a series of ¹H,¹H-COSY and ¹H,¹H-TOCSY experiments along with obtaining ¹H,¹³C-HSQC NMR spectra. These data were fully consistent with the earlier work.⁹ To finally resolve whether the connection between bisulfite and C1’ was via an O or the S atom of bisulfite, we used two methods, (a) Raman spectroscopy, and (b) chemical reactivity.

The Raman spectrum of compound 1 showed an intense peak at 782 cm⁻¹ consistent with a C-S stretch, and this peak was absent in the spectrum for 5. This and other differences between the Raman spectra of 1 and 5 led to the assignment of compound 1 as the S adduct and compound 5 as an O adduct. As further confirmation, we subjected 5 to oxidizing conditions (I₂/H₂O) and found that it resulted in the formation of Ψ; this suggested that the sulfate monoester formed from oxidation could easily ring close to reform the furanose of Ψ upon elimination of sulfate (Scheme 2). Compound 1 did not react with I₂. Finally, we noted that both 1 and 5 were formed as single diastereomers, and we were able to assign their absolute configurations as (S)-1 and (R)-5 based on the close match between experimental ECD spectra and those modeled by TD-DFT calculations. Thus, both 1 and 5 are formed stereospecifically from the native isomer of pseudouridine, providing clues to their mechanisms of formation. Furthermore, calculations led to a model structure of isomer 1 showing a low-energy conformation with H1’-H2’ having a dihedral angle of ~80° in accord with the unusual observation of J_H1’H2’ = 0 Hz.

Scheme 2.

Oxidation of (R)-5 yields an intermediate that ring closes to reform Ψ.

2.2. Proposed Mechanism

Modern organic chemistry textbooks are of little help in understanding the reaction mechanisms responsible for converting pseudouridine, and not other pyrimidines, to C1’ ring-opened bisulfite adducts. Decades ago, bisulfite addition to carbonyl compounds was a routine way of purifying aldehydes and ketones because the bisulfite adducts were often solid salts that could be quantitatively converted back to the carbonyl compound after recrystallization. As advances in chromatography have largely displaced recrystallization, this interesting nucleophile is now absent from textbooks. Nevertheless, the literature has well-documented the conjugate addition of bisulfite to C6 of pyrimidines, and its use in DNA mapping of epigenetic 5mC sites dates to 1992.¹⁰

As a first step, conjugate addition to either π face of pseudouridine should generate a set of four diastereomers resulting from the two new stereogenic centers at C5 and C6; two of these initial adducts would have C5-H and C6-SO₃⁻ oriented trans to one another and be capable of undergoing an E2 elimination to revert to Ψ. The other two would have cis orientations and be slow to revert back to Ψ. We assign compounds 2 and 6 as trans adducts because they reform Ψ upon heating, and 3 and 4 are then the cis-adducts or species derived from them (Scheme 3). For example, the cis adducts are well set up to undergo E2 eliminations leading to ribose ring opening and the formation of Z-alkenes with the newly formed double bond between C1’ of ribose and C5 of uracil. If these alkenes are 3 and 4, then the two final products, (S)-1 and (R)-5, result from excess HSO₃⁻ doing an S_N2’ displacement of the initially adducted bisulfite yielding 1, a C1’ S-adduct; alternatively, a suprafacial migration of sulfite from C6 to C1’ as a [2,3] sigmatropic rearrangement, for which we could find no literature precedent, yields the O-adduct 5. From the four diastereomers initially produced, the two that go on to products 1 and 5 do so by a series of reactions—E2 plus either an S_N2’ or a sigmatropic shift—all of which are stereochemically well defined. Thus, only one diastereomer of 1 and one diastereomer of 5 are formed. We found that the ratio of [1]:[5] was impacted by bisulfite concentration (favoring the S-adduct 1), temperature (favoring the rearranged O-adduct 5), and pH (1:2 ratio at pH 5 and 1:9 at pH 7). As far as we could tell, both ring-opened final products 1 and 5 led to the same outcome during a polymerase bypass experiment (see section 3.1). Nevertheless, it would seem optimal to have principally one product resulting from the bisulfite treatment of pseudouridine.

Scheme 3.

Proposed mechanism for bisulfite reaction with Ψ.

Overall, what does one learn from the study of mechanism? Beyond being intellectually satisfying to the physical organic chemist, the study of reaction mechanism and product distribution should guide the design of protocols to make nucleic acid modification reactions highly specific to the designated target and high yielding to produce a measurable outcome. In the present case, the similarities and differences of bisulfite reactions with cytidine and pseudouridine could be understood, leading to reaction conditions that optimize the ability to target one or both. For example, by simple adjustment of pH, the reaction with C can be greatly diminished without impacting the bisulfite chemistry of Ψ (see section 3.3). In addition, the study of mechanism led to ideas about how to substitute other nucleophiles for capture and biotinylation of RNA modifications; this would have utility in pulling down RNA fragments with low occupancy modifications for amplification and sequencing.

3. CHEMICAL SEQUENCING FOR PSEUDOURIDINE IN RNA

3.1. Deletion Signature from Polymerase Bypass

In a standard chemical sequencing experiment searching for modifications, one compares the sequencing result obtained before and after treatment with a base-specific reagent.¹¹ Typically, two types of outcomes are possible: (1) The reagent may form an adduct that is too large or distorting to be bypassed during reverse transcription (RT) while extending a DNA primer on an RNA template, resulting in an RT stop.¹² PCR amplification of the extended cDNA and Next-gen sequencing reveals the site of the original modification as the 3’ endpoint. (2) The chemistry of the reagent may convert the target base into a new structure that codes like a different base. This is the case for bisulfite sequencing for m⁵C in which C is converted to U but m⁵C continues to code like C during RT.⁵ This second method has the advantage of read-through sequencing so that multiple modifications can be identified on the same strand of RNA. On the other hand, bisulfite sequencing at pH 5 converts such a large number of Cs to Us that sequence alignment with the original RNA becomes a problem.

In the case of pseudouridine, a synthetic RNA oligonucleotide containing a purified mixture of the Ψ-SO₃⁻ adducts at one known location was studied in reverse transcriptase experiments to confirm the behavior of these enzymes as they bypassed pseudouridine adducts in near quantitative yields.² In all conditions studied, the oligomers were transcribed to the end of the template, but the final strands were all n-1 in length (Figure 2a). This was consistent with the original findings of Vahid Khoddami who observed a deletion at every pseudouridine,¹ and we attributed the deletion to the RTase skipping over the unstructured ring-opened Ψ-SO₃⁻ adduct.² Precedent for this is found in the work of Grollman and coworkers who examined polymerase bypass of ring-opened abasic sites that led to a deletion signature.¹³

Figure 2.

Bisulfite treated RNA at (a) pH 5 allows sequencing Ψ via a deletion, m⁵C via no reaction while C deaminates to U, and the bisulfite treatment at (b) pH 7 enables sequencing for Ψ via a deletion signature.

3.2. RBS-seq Identifies Three Modifications on the Same RNA Strand

Before even deciphering the structures of 1 and 5, it was clear that Vahid’s discovery of bisulfite chemistry leading to a sequencing signature for both Ψ and m⁵C would see useful application in sequencing for RNA modifications.¹ In the same reaction, unmodified Cs were converted to Us, and Ψs produced a deletion signature (Figure 2a). Because the reaction conditions involved heating for several hours, a third modification could be sequenced in the same experiment: m¹A. Methylation of N1 of adenosine is an uncommon occurrence in RNA, but it results in a polymerase stop because it blocks the Watson-Crick face of A. Interestingly, m¹A can be readily converted to m⁶A via a Dimroth rearrangement that takes place at the elevated temperatures and higher pH used in the bisulfite desulfonation reaction.¹⁴ Thus, a stop point in the original sequence is converted to a less blocking modification that reads like an A by a polymerase. Overall, the “RNA bisulfite sequencing” protocol that we first published, or “RBS-Seq”, could identify 3 modifications in one experiment.¹ An example of this was demonstrated with tRNA^Gly that has four closely spaced m⁵C modifications amidst one Ψ and one m¹A site.¹

When evaluating whole transcriptomes, RBS-seq identified 486 m⁵C sites and 754 unique sites for Ψ in HeLa RNA.⁹ Previous studies had claimed up to 10,000 m⁵Cs;¹⁵ however, the high number of false positives probably resulted from inadequate denaturation. This is particularly a problem for m⁵C in which the sequencing method relies on near quantitative conversion of all unmethylated Cs, which is very difficult to achieve unless the bisulfite concentration is high, the temperature is relatively high, and the reaction time is long. In contrast, the bisulfite reaction for Ψ is looking for a positive reaction with a small number of sites, i.e., the modification is being positively identified by the chemical reaction rather than negatively identified as in m⁵C sequencing. This makes the bisulfite reaction with Ψ particularly attractive. Furthermore, the fact that the bisulfite modification gives read-through sequencing rather than a polymerase stop is advantageous for finding closely spaced Ψs. Prior to RBS-seq, several labs had adapted the alkylation of Ψ + G by a carbodiimide, CMC, followed by adduct release from G under alkaline conditions to sequence for Ψ.^12,16–18 However, the data from four different studies examining CMC stop points showed very poor overlap,¹⁹ suggesting to us (but unfortunately not to reviewers) that bisulfite sequencing for Ψ might fill a technology gap. Vahid’s discovery was eventually published in 2019,¹ nearly a decade after his initial findings with pseudouridine.

3.3. pH Dependency of Bisulfite Addition

Pseudouridine was among the first modified nucleosides to be identified in RNA,²⁰ and interest in its cellular role has grown in recent years, presenting a need for Ψ sequencing alone. At pH 5, the bisulfite addition reactions yield sequencing changes for C, but not m⁵C, and for Ψ, but not U.¹ To create a Ψ-specific sequencing reaction based on bisulfite, one need only examine the mechanism of the reaction with C. The initial conjugate addition of bisulfite, a bulky sulfur nucleophile, to C is facilitated by protonation of N3, a site with a pK_a of ~4.2, and this increases the electrophilicity of C4 and C6. Thus, m⁵C protocols conduct the initial reaction of bisulfite at pH 4.5–5.0. Simply raising the pH to ≥7 should dramatically inhibit the bisulfite addition to C without impacting Ψ, and indeed that is the case. Using nucleosides, we monitored bisulfite adduction using 2 M NaHSO₃ at 50 °C and found that the rate of reaction for C was nearly undetectable at pH 7, while the rate for Ψ was slowed to about half at pH 7 compared to pH 5.²¹ Nevertheless, there is sufficient reactivity of Ψ at pH 7 to develop highly efficient sequencing protocols for Ψ alone, and precisely that was done recently in both the He laboratory with BID-seq,²² and shortly thereafter in the Yi laboratory where the PRAISE protocol was developed (Figure 2b).²³

3.4. Alternative Nucleophiles

Conjugate addition of nucleophiles to C6 of Ψ works wells for bisulfite, and so we next directed efforts to alternative nucleophiles. Unfortunately, azide, cyanide, and thiocyanate led to only 10–20% yields of adducts to Ψ and C at pH 5.²¹ The ultimate goal of finding an efficient method to biotinylated Ψ specifically using a suitably functionalized nucleophile for addition to C6 of the pyrimidine or for trapping of intermediates 3 and 4 has not yet been successful. An alternative was recently presented by the He laboratory who used metabolic labeling to propargylate N1 of Ψ.²⁴ This can then be used to pull down low occupancy sites in RNA fragments for sequencing.

To summarize, the pH-dependent reaction of bisulfite with either C plus Ψ (pH 5) or Ψ alone (pH 7) leads to predictable DNA sequencing signatures for m⁵C and Ψ after reverse transcription to cDNA. The overall process has two chemical steps followed by reverse transcription, amplification, and library preparation for sequencing. Meanwhile, we began looking for alternatives that would involve direct single-molecule sequencing of strands of nucleic acids to identify positions of base modifications. In 2009, we teamed up with Henry White’s laboratory at the University of Utah to test the power of protein nanopores such as α-hemolysin to report on DNA modifications as single strands were electrophoretically driven through a membrane-embedded pore.²⁵ In this work, we learned that electrical current level, dwell time, and noise could all be useful parameters to evaluate the presence of canonical vs. modified bases in the pore.^25–27 Similar studies elsewhere led to the development of Oxford Nanopore’s MinION technology, a platform we embraced about a decade later.

4. NANOPORE DIRECT RNA SEQUENCING FOR PSEUDOURIDINE USING DWELL-TIME ANALYSIS

4.1. Origin of the Long-Range Helicase Signal

Nanopore direct RNA sequencing is conducted without the introduction of biases during reverse transcription of the RNA to a cDNA and PCR amplification of the cDNA to yield amplicons for NGS. Direct RNA sequencing allows sequencing modifications in their native form without the application of chemicals or antibodies selective for the modifications to raise a signal from the background.^1,22,23 The commercial nanopore sequencer is comprised of a lipid-bilayer embedded protein nanopore sensor with a central constriction zone slightly larger than single-stranded RNA for recording current levels as the RNA passes the pore (Figure 3a).²⁸ Within the central constriction zone resides ~5 nts of RNA that contribute to the current levels, referred to as a k-mer. A second protein required for the success of the system is an ATP-dependent 3′,5′-helicase that delivers the RNA into the nanopore at a rate suitable for recording the current level differences. Successful RNA sequencing with the nanopore requires one round of reverse transcription to yield a DNA:RNA heteroduplex that is the substrate for the helicase feeding RNA into the nanopore while releasing the cDNA back into the solution. We synthesized RNA strands by in vitro transcription (IVT) with Ψ in known sequence contexts separated by >20 nts, a distance greater than the helicase to nanopore proteins, to enable monitoring the signals of a single Ψ compared to a U control in the system.^3,29

Figure 3.

(a) The nanopore direct RNA sequencer is comprised of two protein sensors for detection of RNA modifications. (b) The helicase can detect modifications in the time domain, and the (c) nanopore can detect modifications in the current domain.^3,4 The current levels are deconvoluted using an AI algorithm to generate the base calls.

The sequencing of RNA occurs 3′ to 5′; thus, the first point at which Ψ interacts with the sequencer is in the helicase motor protein (Figure 3b). We found the helicase stalled when translocating through Ψ compared to U.³ A noteworthy point is that sequencing reads are aligned to the base-called data derived from the current level changes in the nanopore; consequently, the signals in the helicase are observed 10–11 nts out of register on the 3′ side relative to the true modification site that is about the distance between the nanopore and helicase. The helicase stalling has been noted in other studies.³⁰ Population histograms of dwell-time data for Ψ in the helicase identified a bimodal distribution, in which one of the modes of the distribution was ~6-times longer in time compared to U, while the other mode of the distribution overlapped with the U population at a lower time (Figure 3b). These data demonstrate that the helicase is a secondary sensor in the sequencer that can be used for identification of Ψ when the other features of the sequencing data are ambiguous in identification of the modification, as described below.

These sequencing experiments analyzed single molecules, and we were effectively conducting a single-molecule enzyme-kinetics experiment on the helicase while it was processing two very similar, yet different, substrates.^3,31 The dwell-time changes led us to question why U produced one time profile, while Ψ produced two time profiles. Three details from the literature help rationalize the findings. (1) Unlike U, which favors the anti conformation of the glycosidic bond, Ψ adopts both the syn and anti conformation with a preference for the syn (Scheme 4a,b).³² (2) Upon reverse transcription, U and Ψ direct the near complete insertion of a dATP opposite in the growing DNA strand,³³ but Ψ can template the dATP in either orientation of the glycosidic bond (Scheme 4b). (3) Lastly, Ψ introduces greater stability to RNA via its additional H-bond donor at N1, and Ψ π stacks in a more stable arrangement than U in duplexes.³⁴ We hypothesize it is these differences between the U:dA vs. Ψ:dA base pairs that influence the helicase activity. In the nanopore sequencer, a variant of a Hel308 helicase is used that employs Phe350 to π stack with the bases as the protein actively pulls the strands apart.^28,35 Pseudouridine in either syn or anti conformations is presented to Phe350 with a different face and projection of the H-bond donor in the active site, in which one orientation is stabilizing compared to U and the other is similar to U to give two different time profiles.

Scheme 4.

Base pairing schemes for (a) U, (b) Ψ, and (c) m¹Ψ in RNA with a dA in complementary DNA strand.

The second point of contact between Ψ and the nanopore sequencer is the nanopore central constriction zone. This is the point at which the current levels change relative to the sequence context to enable sequencing. The current levels are deconvoluted into base calls using an AI algorithm that was trained on canonical nucleotides; as a result, modified nucleotides in RNA will be called as either A, C, G, or U. If the modification alters the current level, this can be reflected in miscalls that reveal the presence of the modification,³⁶ which is a common way to locate RNA modifications in the data.^37–39 In the synthetic RNA that placed the control U or Ψ >20 nts apart in more than 100 contexts, we learned Ψ can impact the current level compared to U with a dependency on the k-mer sequence.^3,29 The location of the maximal current level change can occur in any of the 5-nt k-mers for which Ψ resides yielding a large window (9 nts) for locating the modification. This impact is illustrated in Figure 3c showing the current level difference for Ψ vs. U is at the position 5′ to the modification site. Regarding the base-call data derived from the current levels, Ψ when miscalled is generally identified as a C,^3,37 and the percentage of miscalls range from ~10–100% with a dependency on the sequence context.²⁹ Because Ψ can be written into nearly any sequence context by pseudouridine synthetases,⁴⁰ this represents a challenge for using base-call data to locate and quantify Ψ by nanopore sequencing.

4.2. Identification of Pseudouridine in SARS-CoV-2 mRNA

Our first journey to study Ψ with nanopore direct RNA sequencing data on a biological sample was to apply the consensus data analysis approach on publicly available data for the SARS-CoV-2 vRNA genome.^41,42 The data deposited included a synthetic control without RNA modifications to be used in a comparative analysis to find Ψ while minimizing the false positives inherent in these data. A unique feature of coronavirus replication is that subgenomic RNA strands are produced that code for the abundant structural proteins the virus requires.⁴¹ We looked for Ψ across the subgenomic RNA reads in the available data to find >100 sites of possible modification in the longest subgenomic RNA.^3,38,43 We then restricted the analysis to the U nucleotides to identify significant dwell-time signatures 10–11 nts 3′ to possibly modified Us. The consensus data analysis led to the discovery of 5 sites consistent with Ψ (Figure 4a); moreover, these 5 sites produced the same signatures in all subgenomic RNAs of this virus, which gave us more confidence in their presence. Biochemical evidence for the Ψ sites was demonstrated by treating 100-mer synthetic RNAs with the possible modification site in the center with pseudouridine synthase 1 or 7 (PUS1/PUS7).³ We found both enzymes could convert the suspect U to Ψ to provide biochemical validation of the sequencing results.

Figure 4.

(a) Nanopore direct RNA sequencing using our consensus data analysis identified Ψ in the SARS-CoV-2 vRNA.³ (b) mRNA vaccines to treat SARS-CoV-2 completely replace U with m¹Ψ during their synthesis by (c) IVT.⁴⁴ (d) The sequence-dependent yields of m¹Ψ inserted in RNA generate by IVT in which the RNA polymerase is given an equimolar ratio of UTP and m¹ΨTP.²⁹

4.3. Probing Nucleotide Selection by Polymerases during IVT Synthesis of mRNA Vaccines

Our studies with SARS-CoV-2 found Ψ at specific sites in the vRNA genome. This contrasts with the mRNA vaccines used globally to combat the virus that have every U replaced with stochiometric levels of N1-methylpseudouridine (m¹Ψ; Figure 4b).⁴⁴ Synthetically, this is achieved by replacing UTP with m¹ΨTP during IVT synthesis of the mRNAs. Prior studies found that introduction of Ψ, or better yet m¹Ψ, into therapeutic mRNA strands decreased their immunogenicity, and without the modifications the mRNA vaccines have poor efficacy and are too immunogenetic for broad applications.⁴⁴ Our next studies were inspired by m¹Ψ in mRNA vaccines, and three other points found in the literature. (1) Therapeutic mRNAs are effective with sub-stochiometric levels of RNA modifications,⁴⁵ similar to their presence in vRNA. An unresolved question from this study was whether their IVT-generated, modified mRNAs using mixtures of canonical and non-canonical NTPs during transcription had the desired ratio at each possible site. (2) When m¹ΨTP completely replaces UTP, higher overall RNA polymerase error rates are observed when compared to UTP alone resulting in mRNAs with greater sequence errors.⁴⁶ (3) Concerns of long-term exposure to highly modified mRNA strands on health have been expressed.⁴⁷ These reports led us to ask how the nanopore sequencer responds to m¹Ψ in RNA, and then can we use this knowledge to monitor IVT generated RNAs with a mixture of UTP and m¹ΨTP to determine whether the polymerase biased the mixture in the RNA polymer?

Synthetic RNA made with m¹Ψ instead of Ψ in >100 5-nt k-mer sequence contexts described previously were then sequenced with the nanopore.²⁹ The results of this analysis found m¹Ψ signals were similar to those for Ψ, in which the base miscall analysis, currents, and dwell times were highly sequence context dependent. With the knowledge of sequencing signatures for m¹Ψ compared to U in all these sequence contexts, IVT reactions were conducted with a mixture of UTP and m¹ΨTP in a 1:1 ratio. Following along the approach of mRNA vaccine synthesis,⁴⁴ we used T7 RNA polymerase for synthesizing the RNA strands with the mixture of U and m¹Ψ; moreover, the template DNA allowed the study of nearly all adjacent sequence contexts in which m¹Ψ resides in the Pfizer and Moderna vaccines (i.e., 5′-VXV-3′ and 5′-VXXV-3′; where V = A, C, or G and X = U or m¹Ψ; Figure 4c). The nanopore direct RNA sequencing results when analyzed quantitatively found T7 RNA polymerase installed U and m¹Ψ in the mRNA with ratios strongly dependent on the adjacent sequence context, with UTP always inserted at a higher frequency (Figure 4d). This finding suggests that if therapeutic mRNAs are made with a mixture of U and m¹Ψ the ratio of the two nucleotides will vary across the polymer depending on the sequence context.

A chemical explanation for why UTP outcompeted m¹ΨTP is best understood using the syn vs. anti glycosidic bond conformation preferences for the two nucleotides. For UTP, the base favors the anti conformation, and when entering the active site of the polymerase opposite a templating dA in the DNA, a catalytically competent structure is formed (Scheme 4a). On the other hand, m¹ΨTP favors the syn conformation that projects the N1-methyl group toward the templating dA to block H-bonding resulting in no extension of the strand (Scheme 4c).³² Regarding the sequence context dependency in the m¹Ψ yields, we hypothesize active site residues in T7 RNA polymerase may play a role in the selection. To study this we repeated the experiments using the SP6 RNA polymerase that differs at many amino acids, in which a key difference is at position 644 where a leucine residue replaces T7’s phenylalanine.⁴⁸ In T7, the phenylalanine π stacks with templating duplex DNA, while in SP6 the leucine cannot participate in this interaction. When SP6 was the RNA polymerase for UTP/m¹ΨTP competition, the nanopore sequencing results identified a more balanced incorporation of m¹Ψ across the sequence contexts. This observation suggests a role for the active site amino acid residues in the polymerase, and likely leucine 644 in NTP selection. For readers more interested in these findings and the other experiments performed to better understand the NTP selection, we refer you to the original publication,²⁹ as the details are beyond this scope of this Account.

4.4. Extension to 17 Modifications in Ribosomal RNA

With our understanding of nanopore direct RNA sequencing for Ψ and m¹Ψ employing base miscall, ionic current, and dwell-time features of the data revealing the presence of modifications, we set our sights on sequencing a greater variety of RNA modifications. The E. coli rRNA strands provide a test case for this question because there are 36 well-established modification sites in the 16S and 23S rRNAs with 17 different chemical structures.⁴ These RNAs are modified with methyltransferases, isomerases, an oxidase, or a reductase to generate the 17 different structures (see Figure 5a for acronyms and locations of the modifications). The rRNAs were extracted from stationary phase grown E. coli that have the modifications written at high occupancy based on mass spectrometry analysis.⁴⁹ Nanopore sequencing of the rRNAs enabled cataloging the base miscalls, ionic currents, and helicase dwell times for each modification (Figure 5a). Use of writer knockouts provided validation for assigning many of the signals and sequencing synthetic RNAs provided additional validations. There was a cluster of Ψ residues in the 23S rRNA and m⁶₂A residues in the 16S rRNA for which the ionic current levels for each of the modifications could not be definitively determined because their respective writer enzymes were the same (Figure 5a).

Figure 5.

(a) Nanopore direct RNA sequencing examining dwell time, current level, and base miscalls for the 17 chemical structures located at 36 different sites in the E. coli rRNA strands.⁴ (b) Changes in E. coli rRNA modifications upon exposure to stress that were determined with nanopore sequencing.⁴ The pH 7 bisulfite reaction to yield Ψ adducts increases the (c) indel frequency at known modified sites, and (d) shows no change in base calls for U/C sequence variations that give similar signals as Ψ in the nanopore data.⁵¹

Using this knowledge, we grow the E. coli under metabolic, heat shock, or cold shock stress and then profiled the modifications in a semi-quantitative fashion to reveal those that changed. The plot in Figure 5b illustrates the statistically significant differences color coded for the E. coli exposed to metabolic or cold-shock stress compared to those not experiencing the stressor. Notable examples are many of the Ψ and m²G sites, as well as the hypermodified 16S m⁴C_m1402 that decreased under all stressors applied, with the exception of a single Ψ site that increased with metabolic stress (Figure 5b). In general, the changes tracked with previously established alterations in the writer protein expression levels in E. coli under the stress conditions studied.⁵⁰ Additionally, ho⁵C decreased under thermal stress that causes oxidative stress, likely resulting from the extreme sensitivity of this oxidized base to further oxidation. These studies demonstrate it is possible to monitor many RNA modifications with the nanopore, but accurate quantification is still challenging to achieve.

5. NANOPORE SEQUENCING OF CHEMICALLY TAGGED PSEUDOURIDINE

5.1. Sequencing Bisulfite Adducts with Nanopore Technology

In our final study on this topic, we married the facile and clean pH 7 bisulfite reaction with Ψ to form a stable ribose adduct for nanopore direct RNA sequencing.⁵¹ We reasoned that the size and charge of the adduct would dominate the signals in the nanopore and minimize or abolish the sequence dependency in the signatures. By producing a more coherent signal for Ψ via the adduct, quantification of these sites in nanopore data would be more easily achieved. The BID-Seq protocol²² was used on the E. coli rRNA to study the 10 known Ψ sites. The base-calling error analysis at known Ψ sites before and after the bisulfite reaction identified that the adduct generates insertion-deletion (indels) errors in the sequencing reads; however, the indel frequencies for the three Ψ sites shown in Figure 5c identify the error is still sequence-context dependent. Further analysis of the results found the pH 7 bisulfite reaction targeting Ψ had two benefits in nanopore sequencing. (1) The adduct allowed detection of Ψ down to levels of ~20% that are below the levels found for unreacted Ψ. (2) The E. coli genome codes for 7 rRNA operons that have 62 known sequence variations, of which 19 are U/C variants. The challenge with U/C sequence variants is they give base-calling signatures that masquerade as Ψ, and therefore, a test to determine whether a suspect site is a real Ψ or a U/C sequence variation is to sequence before and after the pH 7 bisulfite reaction. In Figure 5d are three representative U/C sequence variation sites that do not change after the reaction, as expected. Utility of site-specific chemistry on RNA modifications can aid in interpretation of nanopore direct RNA sequencing data.

6. SUMMARY AND OUTLOOK

Pseudouridine is the most common epitranscriptomic modification found in all RNA types that is written by 13 pseudouridine synthases in humans and 11 in E. coli.^12,52 The occupancy of Ψ across the transcriptome is organism, cell-type, and stress dependent, suggesting this U isomer has significant role in cellular processes.^{4,12,22,23,37} Determination of these roles will require accurate and quantitative sequencing of RNA, in which nanopore direct RNA sequencing is a candidate method to achieve this goal.^37,53 Our work identified that nanopore signatures for Ψ are highly sequence-context dependent and that using a consensus of nanopore current, helicase dwell time, and base miscalling is an approach to sequence this U isomer.³ The pH 7 bisulfite reaction to yield Ψ adducts results in detection of adducts via a deletion signature in traditional sequencing after reverse transcription,^1,22,23 and it is found as an indel in nanopore direct RNA sequencing.⁵¹ In the nanopore data, the indel results from the current levels being poorly recognized by the software. Future developments in nanopore data analysis software will allow this modified nucleotide and likely others to be directly sequenced. The choice of using bisulfite sequencing for Ψ by either traditional cDNA sequencing or nanopore sequencing will be most dependent on the amount of sample available. Traditional cDNA sequencing requires PCR and will be most applicable for small samples, while nanopore direct RNA sequencing can provide a more comprehensive picture of RNA modifications when larger sample amounts are available.

Biographies

Cynthia J. Burrows is a distinguished professor of chemistry and the Thatcher chair of biological chemistry at the University of Utah where she often teaches a course in nucleic acid chemistry. In addition to RNA modifications, her research interests include DNA modifications resulting from oxidative stress and the enzymes that process them. She is a member of the U.S. National Academy of Sciences.

Aaron M. Fleming is a research associate professor of chemistry in the Burrows laboratory. His research interests include understanding the chemistry and biology of DNA and RNA chemical modifications and employing nanopore sequencing to study these modifications.