Pseudouridines have context-dependent mutation and stop rates in high-throughput sequencing

ABSTRACT

The abundant RNA modification pseudouridine (Ψ) has been mapped transcriptome-wide by chemically modifying pseudouridines with carbodiimide and detecting the resulting reverse transcription stops in high-throughput sequencing. However, these methods have limited sensitivity and specificity, in part due to the use of reverse transcription stops. We sought to use mutations rather than just stops in sequencing data to identify pseudouridine sites. Here, we identify reverse transcription conditions that allow read-through of carbodiimide-modified pseudouridine (CMC-Ψ), and we show that pseudouridines in carbodiimide-treated human ribosomal RNA have context-dependent mutation and stop rates in high-throughput sequencing libraries prepared under these conditions. Furthermore, accounting for the context-dependence of mutation and stop rates can enhance the detection of pseudouridine sites. Similar approaches could contribute to the sequencing-based detection of many RNA modifications.

Introduction

Pseudouridine (Ψ) is the most abundant of over a hundred different modifications that occur in cellular RNAs [1,2]. The RNA-guided box H/ACA small nucleolar ribonucleoproteins and RNA-independent pseudouridine synthase enzymes post-transcriptionally isomerize uridine (U) to generate pseudouridine at specific sites in noncoding and coding RNAs, including ribosomal RNA (rRNA), transfer RNA (tRNA), small nuclear RNA (snRNA), and messenger RNA (mRNA) [1–4]. Pseudouridines in noncoding RNAs are conserved and cluster in functionally important regions, such as the peptidyl transferase center in rRNAs and the branch-site recognition sequence in U2 snRNA [3]. By increasing phosphodiester backbone rigidity and base stacking [5–7], pseudouridines impact intramolecular and intermolecular RNA–RNA interactions, including rRNA folding, tRNA binding to the ribosome, and pre-mRNA–snRNA base-pairing [8–10]. Pseudouridines in mRNAs are also conserved [11,12], can alter decoding by the ribosome [1,13,14], and have been proposed to stabilize modified mRNAs during heat shock [11]. While pseudouridine modifications are thought to be irreversible [15], pseudouridylation at sites in both noncoding and coding RNAs can be substoichiometric [4,16,17], and modifications at some sites are induced in response to changes in growth conditions or stress [11,18,19].

Pseudouridine sites in cellular RNAs can be detected through the selective chemical modification of pseudouridines with N-cyclohexyl N′-(2-morpholinoethyl) carbodiimide (CMC) [20]. CMC reacts with pseudouridine, uridine, and guanosine, but only the CMC-modified N3 of pseudouridine is resistant to alkaline hydrolysis (Fig. 1A). Following CMC treatment, pseudouridine sites can be identified by primer extension, since reverse transcription terminates one nucleotide 3′ to CMC-modified pseudouridine (CMC-Ψ). The combination of this method with high-throughput sequencing revealed hundreds of pseudouridine sites in mRNAs and noncoding RNAs previously thought to lack pseudouridylation [11,12,19]. The sensitivity of sequencing-based pseudouridine detection was further improved by using a chemically synthesized derivative of CMC that allowed the biotinylation and enrichment of pseudouridine-modified transcripts prior to reverse transcription [21]. These studies also identified several novel pseudouridine sites in rRNA, of which three were validated through direct detection of the pseudouridine nucleoside using site-specific cleavage and radioactive-labeling followed by ligation-assisted extraction and thin-layer chromatography (SCARLET) [19,21,22]. Together, these studies greatly expanded the scope of known pseudouridine modifications in eukaryotic RNAs.

Figure 1.

Reverse transcription through CMC-modified Ψ. (A) Chemical structures of Ψ and CMC-modified Ψ (CMC-Ψ). (B) Reverse transcription of a synthetic RNA oligo (Oligo Ψa) containing Ψ or CMC-Ψ with AMV RT or HIV RT. RNA oligo sequence: 5′- UACACUCAGXUCGGACUAAAGCUGCUC (X = Ψ or CMC-Ψ). (C) Quantification of Ψ or CMC-Ψ read-through by different reverse transcriptase enzymes under varying divalent cation conditions.

Nonetheless, sequencing-based pseudouridine detection methods still have ample room for improvement. Only a small fraction of the newly discovered pseudouridine sites were reproducibly detected by more than one pseudouridine sequencing study [23]. In addition, quantitative mass spectrometry estimates the Ψ/U ratio in mammalian mRNAs at ∼0.2–0.4%, which would correspond to thousands of pseudouridine sites, or several times more sites than have been identified even with the pre-enrichment of pseudouridylated RNAs [21]. One element that limits the specificity and sensitivity of current methods is the use of reverse transcription stops to map pseudouridine sites: stops can also result from RNA structure or degradation, and pseudouridine sites that are close to another downstream pseudouridine or to the 3′ end of an RNA molecule are less likely to be detected. An alternative approach would be to detect mutations introduced through misincorporation during reverse transcription, rather than detecting stops. Mutation-based approaches have been used to map RNA modifications transcriptome-wide [24-27], and the combined analysis of both stop and mutation rates can improve the detection of modification sites as compared to the analysis of either stop or mutation rate alone [28]. Since sequence context can influence the identities of misincorporated bases [28], accounting for the surrounding nucleotide sequence could further improve the detection of RNA modification sites.

In this work, we show that pseudouridine modifications in CMC-treated human ribosomal RNA have context-dependent mutation and stop rates in high-throughput sequencing libraries prepared under specific reverse transcription conditions. By testing different reverse transcriptase enzymes and divalent cations, we found conditions in which reverse transcriptase reads through CMC-modified pseudouridine in a synthetic RNA oligo. We then used these reverse transcription conditions to prepare high-throughput sequencing libraries from CMC-treated and mock-treated human rRNA. The proportions of stops and mutations observed at pseudouridine sites in CMC-treated RNA varied depending on the identity of the nucleotide 3′ to pseudouridine. By defining different stop and mutation thresholds for different sequence contexts, we were able to improve the sensitivity and specificity of pseudouridine detection relative to context-independent thresholds. Thus, accounting for context-dependent stop and mutation rates can enhance the detection of RNA modification sites in high-throughput sequencing data.

Results

Reverse transcription through CMC-modified pseudouridine in an RNA oligo

We set out to find conditions in which reverse transcription can bypass CMC-modified pseudouridine (CMC-Ψ). We synthesized a 27-nucleotide RNA oligo containing a single pseudouridine, treated it with CMC to obtain an oligo containing a single CMC-Ψ, and reverse transcribed both the CMC-Ψ-containing and untreated Ψ-containing oligos under various conditions. Avian myeloblastosis virus (AMV) reverse transcriptase (RT) synthesized full-length complementary DNA (cDNA) from the Ψ-containing oligo template with high efficiency (Fig. 1B–C), as expected given that pseudouridine and uridine are essentially indistinguishable by RT enzymes. When the CMC-Ψ-containing oligo was used as the template, most of the cDNAs terminated one nucleotide 3′ to the CMC-Ψ. When varying proportions of the Ψ-containing and CMC-Ψ-containing oligo templates were reverse transcribed, the quantity of truncated cDNAs scaled with the proportion of CMC-Ψ-containing template. Since human immunodeficiency virus (HIV) RT has previously been shown to read through bulky adducts in template RNAs [29]. we tested whether HIV RT could read through CMC-modified pseudouridine. Indeed, HIV RT produced nearly equal amounts of full-length cDNA from Ψ-containing and CMC-Ψ-containing oligo templates, corresponding to ∼84% read-through of CMC-Ψ (Fig. 1C). Since the manganese divalent cation (Mn²⁺) has previously been used to enhance read-through of bulky 2′-O-adducts by SuperScript II RT [24], we investigated the effect of different divalent cations on CMC-Ψ read-through. While SuperScript III RT behaved similarly to AMV RT under standard reaction conditions with divalent cation Mg²⁺, replacing Mg²⁺ with at least 3 mM Mn²⁺ (with 0.5 mM of each dNTP) led to ∼88% read-through of CMC-Ψ (Fig. 1C and S1). Thus, we identified two conditions that facilitate reverse transcription through CMC-Ψ: (1) reverse transcription with HIV RT and (2) reverse transcription with SuperScript III RT in the presence of 3 mM Mn²⁺.

High-throughput sequencing of pseudouridine sites in human ribosomal RNA

Since the Watson-Crick face of pseudouridine is indistinguishable from that of uridine, reverse transcriptases incorporate an adenosine (A) nucleotide opposite of Ψ. However, CMC modifies the Watson-Crick face of pseudouridine (Fig. 1A), so reverse transcriptases that read through CMC-Ψ might incorporate T, C, or G rather than A opposite of CMC-Ψ. If misincorporation occurs, then CMC-modified pseudouridine sites can be identified as mutated U's in RNA sequencing data. To test this possibility, we prepared cDNA libraries of human rRNA using reverse transcription conditions that allow CMC-Ψ read-through (Fig. 2A). We isolated total RNA from human embryonic kidney 293T cells (HEK293T) and size-selected for RNA molecules over 200 nucleotides long to enrich for rRNA. After fragmentation, a control RNA oligo with one partially modified site (50% Ψ and 50% U) was added. Next, the RNA was split into two parts: one part was treated with CMC followed by reversal at alkaline pH (CMC+), while the other part was mock-treated in buffers lacking CMC (CMC−). Following end repair and 3′ adapter ligation, the CMC+ and CMC− samples were each split into three equal parts that were reverse transcribed with (1) HIV RT, (2) SuperScript III RT with Mn²⁺ (SIII RT, Mn²⁺) or (3) SuperScript III RT with Mg²⁺ (SIII RT, Mg²⁺). For the CMC-treated sample (CMC+), reverse transcription conditions (1) and (2) were expected to generate cDNAs with misincorporations at CMC-Ψ sites, while condition (3) was expected to generate cDNAs terminating one nucleotide 3′ to CMC-Ψ. The cDNAs generated from all six conditions (CMC+/−, with 3 reverse transcription conditions each) were gel-purified, amplified, and sequenced.

Figure 2.

High-throughput sequencing of Ψ sites in human rRNA. (A) Sequencing workflow for detection and analysis of Ψ sites in human rRNA. Reverse transcription through CMC-modified Ψ sites with HIV. RT or SuperScript III (SIII) RT can produce read-through cDNAs with or without mismatches, or truncated cDNAs. (B) Mutation rates around Ψ1004 in18S rRNA, which has a +1 G context, in libraries prepared with HIV RT, SuperScript III RT with Mn²⁺, or SuperScript III RT with Mg²⁺. (C) Stop rates around Ψ1445 in 18S rRNA, which has a +1 A context, in libraries prepared with HIV RT, SuperScript III RT with Mn²⁺, or. SuperScript III RT with Mg²⁺.

We evaluated the mutation and stop rates at known pseudouridine sites in CMC-treated and mock-treated human 18S and 28S rRNAs [30]. To minimize the background mutation rate, one nucleotide was trimmed from either end of each read. Stop count assignments were also shifted over by one nucleotide position to account for this end-trimming step. When we took the difference (Δ) between mutation or stop rates in the CMC+ and CMC− samples, we found that known pseudouridine sites in rRNA tended to have higher Δmutation rates and lower Δstop rates in the sequencing libraries prepared using HIV RT or SIII RT, Mn²⁺ than in libraries prepared using SIII RT, Mg²⁺ (Fig. S2). Although these general trends were consistent with our expectations, the Δmutation rates at known pseudouridine sites were unexpectedly low, with a median Δmutation rate < 0.04 even in the libraries prepared under reverse transcription conditions that favored CMC-Ψ read-through. Moreover, the Δstop rates in libraries prepared using HIV RT or SIII RT, Mn²⁺ were only slightly lower than the Δstop rate in libraries prepared using SIII RT, Mg²⁺, suggesting that CMC-Ψ read-through during library preparation was less efficient than the CMC-Ψ read-through we observed during reverse transcription of our model oligo (Fig. 1B–C). To test whether CMC-Ψ read-through led to skipped or added nucleotides, we re-aligned our sequencing data to allow deletions and insertions. Deletion rates were < 0.03 at all but 4 out of 36 pseudouridine sites in the 18S rRNA, whereas insertion rates occurred at similarly low levels in the libraries prepared from CMC-treated and mock-treated rRNA. Therefore, we did not account for deletions or insertions in our subsequent analysis. While lower than expected, Δmutation rates in the libraries prepared using HIV RT or SIII RT, Mn²⁺ were sufficient to identify certain pseudouridine sites, such as Ψ1004 in 18S rRNA, which has a G at the +1 position, i.e. one nucleotide 3′ to the pseudouridine (Fig. 2B and S3A). In contrast, other pseudouridine sites, such as Ψ1445 in 18S rRNA, which has an A at the +1 position, were best identified based on Δstop rates in libraries prepared using any of the three reverse transcription conditions (Fig. 2C and S3B).

Context-dependent mutation and stop rates at pseudouridine sites

We examined the effect of sequence context on mutation and stop rates at pseudouridine sites in CMC-treated and mock-treated rRNA. We limited this analysis to the sequencing libraries prepared with HIV RT, which had higher median Δmutation rates at known pseudouridine sites than libraries prepared with SIII RT, Mn²⁺ or SIII RT, Mg²⁺ (Fig. S2). We found that the nucleotide 3′ to pseudouridine in the RNA template (the +1 nucleotide) influenced the Δmutation and Δstop rates at known pseudouridine sites, but not at unmodified uridine sites, in human 18S and 28S rRNAs (Fig. 3A–C). Known pseudouridines with G at the +1 position (+1 G) tended to have high Δmutation rates and low Δstop rates, whereas known pseudouridines with A at the +1 position (+1 A) tended to have high Δstop rates and low Δmutation rates (Fig. 3A and 3C). The Δmutation and Δstop rates at known pseudouridines with C or U at the +1 position (+1 C or +1 U) tended to follow the same patterns as the rates at known pseudouridines with +1 G or +1 A, respectively, but these patterns were less pronounced (Fig. 3A). Overall, known pseudouridines with +1 C or +1 U could have either elevated Δmutation rates or elevated Δstop rates (Fig. 3A), resulting in intermediate median Δmutation and Δstop rates (Fig. 3C). The trend in context-dependent Δmutation rates (+1 G > +1 U ∼ +1 C > +1 A) was not fully explained by the trend in Δstop rates, since this pattern remained when stops were not counted (median Δmutation rates without counting stops: 0.104 for +1 G, 0.056 for +1 U, 0.037 for +1 C, and 0.023 for +1 A). Next, we examined the impact of the +1 nucleotide on the identity of the bases that were misincorporated opposite of CMC-Ψ during reverse transcription, or the mutation signature (Fig. S4). Regardless of the +1 nucleotide, known pseudouridine sites in rRNA had their highest Δmutation rate to C, and a higher Δmutation rate to A than to G, indicating that G tended to be misincorporated most often, and U was misincorporated more often than C. Thus, the total Δmutation and Δstop rates at known pseudouridine sites in rRNA varied depending on the +1 nucleotide, but the mutation signature was not strongly dependent on the +1 nucleotide.

Figure 3.

Context-dependent mutation and stop rates at Ψ sites. ΔMutation and Δstop rates at all known Ψ sites (A) and at all unmodified U's (B) in 18S and 28S rRNA, in RNA libraries prepared with HIV RT. Dashed and dotted lines represent the context-dependent thresholds used to identify Ψ sites (+1 A: Δstop > 0.2; +1 C: ΔMI > 0.15 OR Δmutation > 0.03; +1 U: Δstop > 0.2 OR Δmutation > 0.04; +1 G: ΔMI > 0.09 AND Δmutation > 0.03). Color and shape specifies the +1 nucleotide context. Mismatch index (MI) = mutation rate + stop rate. ΔRate = rate in CMC-treated sample (CMC+) − rate in mock-treated sample (CMC−). (C) Context-dependent mutation and stop rates. Box-and-whisker plots of Δmutation and Δstop rates at known Ψ sites with different +1 nucleotide contexts, in libraries prepared with HIV RT. (D) Receiver operating characteristic (ROC) curves for Ψ site detection in 18S and 28S rRNA. For context-dependent thresholds, Δmutation, Δstop, and/or Δmismatch index (MI) thresholds vary depending on the +1 nucleotide. The thresholds used to construct this curve were selected to maximize sensitivity and specificity. The red X marks the context-dependent thresholds shown in Fig. 3A–B. Context-dependent thresholds can increase both sensitivity (shift up) and specificity (shift left) of Ψ site detection, as compared to Δmutation, Δstop, or ΔMI thresholds that do not account for +1 nucleotide context.

Previous pseudouridine sequencing studies have relied on stop rates or other stop-based metrics to set thresholds for calling pseudouridine sites [11,12,19,21,23]. Using our rRNA sequencing data, we compared the performance of Δstop rate thresholds to the performance of Δmutation rate thresholds and of thresholds in a combined metric, Δmismatch index (MI = stop rate + mutation rate). Since Δmutation and Δstop rates were dependent on +1 nucleotide context, we also tested context-dependent thresholds, in which different Δmutation and Δstop rate thresholds were established for different +1 nucleotide contexts. We found that Δmutation rate thresholds and Δstop rate thresholds performed similarly, Δmismatch index thresholds performed better, and context-dependent thresholds had the highest sensitivity and specificity for pseudouridine sites in human rRNA (Fig. 3D).

Discussion

In this study, we identified reverse transcription conditions that favor CMC-Ψ read-through, and we showed that pseudouridines in CMC-treated rRNA have context-dependent mutation and stop rates that can be used to improve pseudouridine site detection in high-throughput sequencing. Using either a different reverse transcriptase enzyme, HIV RT, or a different divalent cation with SuperScript III RT, Mn²⁺, led to over 80% read-through of CMC-Ψ in a synthetic RNA oligo (Fig. 1 and S1). These reverse transcription conditions also decreased the stop rate and increased the mutation rate at pseudouridine sites in a CMC-treated human rRNA sequencing library (Fig. S2), though not to the same degree as expected based on the efficient CMC-Ψ read-through observed with the model oligo. This discrepancy could be due to changes in reverse transcriptase efficiency and processivity in the setting of extracted cellular RNA, as the reverse transcription reaction is highly sensitive to reaction parameters including RNA concentration [31,32]. The low mutation rate at pseudouridine sites in CMC-treated rRNA could also be due to the incorporation of A opposite of CMC-Ψ, which would produce neither a stop nor a mutation. In fact, the low Δmutation rate and the mutation signature observed at known pseudouridine sites in rRNA (Figs. S2 and S4) follow the same patterns as non-templated synthesis, in which HIV RT prefers to add A or G, with a lower preference for T [33], suggesting that HIV RT conducts non-templated synthesis upon encountering CMC-Ψ.

Although the stop rate was higher and the mutation rate lower than expected based on our results with the model oligo, these sequencing results allowed us to incorporate both stop and mutation information in our analysis. Consistent with previous results with a different RNA modification [28], the combined analysis of stop and mutation rates enhanced the prediction of modification sites relative to evaluating stops or mutations alone (Fig. 3D). Previous work has also shown that the nucleotide 3′ to an RNA modification site can affect its mutation signature [28]. In our sequencing data, the +1 nucleotide did not influence the mutation signature at pseudouridine sites in CMC-treated rRNA, but it did significantly impact the stop and mutation rates. Known pseudouridines with +1 G or, to a lesser extent, +1 C, tended to have higher Δmutation rates and lower Δstop rates than average, whereas known pseudouridines with +1 A or, to a lesser extent, +1 U, tended to have higher Δstop rates and lower Δmutation rates than average (Fig. 3A–C). These patterns can be explained by the higher stability of G-ribonucleotide–C-deoxyribonucleotide (rG–dC) and rC–dG base pairs relative to rA–dT and rU–dA base pairs [34], since a more stable base pair between the +1 nucleotide and the 3′ end of the cDNA could prevent fraying of the RNA–cDNA hybrid and thereby increase the efficiency of cDNA extension when the reverse transcriptase encounters CMC-Ψ. The dependence of the measured stop rates on +1 nucleotide context could also result from 3′-end nucleotide biases in the circularization step. The observed stop rates follow the pattern +1 A > +1 U > +1 C > +1 G. Reverse transcription stops at pseudouridine sites with these +1 positions in the template RNA would lead to truncated cDNAs with 3′-end nucleotides T, A, G, and C, respectively. Therefore, the observed pattern in stop rates matches the 3′-end nucleotide preferences of CircLigase I (3′ T > 3′ A > 3′ G, with no detectable ligation for 3′ C, according to Epicentre). Although the end-base preferences of CircLigase II have not been reported, CircLigase II is an adenylated form of CircLigase I and therefore mostly likely has similar sequence preferences. However, the trend in mutation rates (+1 G > +1 U ∼ +1 C > +1 A) was observed even when stops were not counted, which cannot be explained by differences in either RNA–cDNA hybrid stability or circularization efficiency.

Based on our observations, we created simple and interpretable context-dependent thresholds for the identification of pseudouridine sites based on high-throughput sequencing of CMC-treated and mock-treated RNA. These context-dependent thresholds required that putative pseudouridines with +1 G have high Δmutation rates, putative pseudouridines with +1 A have high Δstop rates, and putative pseudouridines with +1 C or +1 U have either high Δmutation rates or high Δstop rates, in order to be identified as pseudouridine sites (Fig. 3A). These context-dependent thresholds improved the detection of pseudouridine sites relative to context-independent thresholds (Fig. 3D). Given the sensitivity of the reverse transcription reaction to various reaction parameters [31,32], the trends in mutation and stop rates at pseudouridine sites in CMC-treated human rRNA, which is highly structured and modified, might differ from the patterns in other RNAs such as mRNA. Nonetheless, the context-dependent patterns observed in rRNA provide a reasonable starting point for the detection of pseudouridines in high-throughput sequencing of other CMC-treated RNA samples. Moreover, regardless of the specific mutation and stop rates, accounting for context-dependence of mutation and stop rates by using context-dependent thresholds could increase the sensitivity and specificity of pseudouridine detection. Thus, context-dependent mutation and stop rates provide valuable information that may enhance the detection of pseudouridine or other RNA modifications in high-throughput sequencing.

Materials and methods

RNA oligo synthesis and CMC modification

Oligonucleotide synthesis was performed on an Expedite Nucleic Acid Synthesis System using standard RNA synthesis conditions on a 1-μmol scale. Controlled Pore Glass (CPG) supports and A, C, G, and T phosphoramidites were purchased from ChemGene. Pseudouridine phosphoramidite was purchased from Glen Research. The terminal 4,4′-dimethoxytrityl (DMTr) protecting group was removed from the oligonucleotides by using the DMTr-off mode. The resin containing the oligos was transferred to a 2-mL vial, and a mixture of 0.3 mL ethanol and 0.9 mL 30% ammonium hydroxide was added. The vial was then incubated at 55°C for 4 hours to remove all the base-labile protecting groups. Once cooled to room temperature, the supernatant was transferred to a 1.5-mL tube and dried in a SpeedVac centrifugal vacuum concentrator. A mixture of 100 µL dimethyl sulfoxide (DMSO) and 125 µL hydrogen fluoride triethylamine (Sigma Aldrich) was added to the tube, and the tube was incubated in a 65°C water bath for 2.5 hours. After cooling to room temperature, 22.5 µL 3 M sodium acetate (pH 5.3) was added, and the mixture was vortexed. Next, 1 mL of n-butanol was added, and the mixture was kept at −80°C overnight. After spinning at maximum speed for 25 minutes at 4°C, the pellet was washed with 1 mL of 70% v/v ethanol and spun again. The pellet was then dissolved in 1 mL of nuclease-free water and purified by C18 reverse-phase high-performance liquid chromatography (HPLC) with the applied buffer as 0–20% acetonitrile in 0.1 M triethylammonium acetate in water. The fractions were checked for purity by analytical HPLC and matrix-assisted laser desorption/ionization mass spectrometry (MALDI-MS). The purified oligonucleotides were concentrated in a SpeedVac centrifugal vacuum concentrator. The sequences of the synthesized RNA oligos were as follows:

Oligo Ψa: 5′-UACACUCAGΨUCGGACUAAAGCUGCUC

Oligo Ua: 5′-UACACUCAGUUCGGACUAAAGCUGCUC

To prepare the CMC-Ψ-containing RNA oligo, 10 µg of Oligo Ψa in 12 µL water were combined with 24 µL of 50 mM Tris–Cl (pH 8.3), 4 mM EDTA, 7 M urea (TEU buffer) and 4 µL of 1 M N-cyclohexyl N′-(2-morpholinoethyl) carbodiimide (CMC) freshly dissolved in TEU buffer, for a final concentration of 0.1 M CMC. The mixture was incubated at 30°C overnight (16 hours), and then the RNA was purified with an Oligo Clean & Concentrator column (Zymo Research, D4061) and eluted in 20 µL water. Next, 40 µL of 50 mM sodium carbonate and 2 mM EDTA (pH 10.4) were added, and the 60-µL reaction was incubated at 45°C for 2 hours or at 37°C for 4 hours. Analytical HPLC and MALDI-MS showed that almost all CMC on U's and G's were removed, while ∼80% of CMC-Ψ adducts remained.

Purification, radiolabeling, and phosphorylation of DNA primers and DNA ladder

The following DNA primers were purchased from Integrated DNA Technologies (IDT):

Primer A: 5′-GAGCAGCTTTAG

Primer B: 5′-GATCGTCGGACTGTAGAACTAGACGTGTGCTCTTCCGATCT

Primer 18S-831: 5′-GTATCCAGGCGGCTCGGGCC

Primer 28S-3755: 5′-GATGACGAGGCATTTGGCTACC

Illumina multiplex primer:

5′-AATGATACGGCGACCACCGAGATCTACACGTTCAGAGTTCTACAGTCCGACGATC

Barcoded primer:

5′-CAAGCAGAAGACGGCATACGAGAT[Barcode]GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCT

Primers A, B, 18S-831, and 28S-3755, and the Illumina multiplex and Barcoded primers, were purified on a gel containing 10% acrylamide:bisacrylamide (29:1), 7 M urea, 89 mM Tris–borate (pH 8.3), and 2 mM Na₂EDTA (ethylenediaminetetraacetic acid). DNA was excised from the gel by UV shadowing and eluted in 50 mM potassium acetate and 200 mM KCl (pH 7.5) by the crush-and-soak method. Eluted RNA was precipitated in ethanol, and then resuspended and stored in water at −20°C. Primers A, 18S-831, and 28S-3755 were 5′ ³²P-labeled with 20 pmol of 6000 Ci/mmol γ-³²P-ATP per 100 pmol of primer, using 10 U of T4 polynucleotide kinase (T4 PNK, New England BioLabs, M0201L) in 70 mM Tris-Cl (pH 7.6), 10 mM MgCl₂, and 5 mM DTT (1 × PNK buffer) in a total volume of 10 µL at 37°C for 30–60 minutes, and then extracted once with phenol:chloroform (3:1), ethanol precipitated, and resuspended to a concentration of 2 µM RNA in water. Primer B was 5′ ³²P-labeled with 13.2 pmol of 6000 Ci/mmol γ-³²P-ATP per 300 pmol of primer, using 10 U of T4 PNK in 1 × PNK buffer in a total volume of 20 µL at 37°C for 20–30 minutes. Next, 1.2 nmol (4 molar equivalents) of cold ATP and an additional 10 U of T4 PNK were added, and the reaction was incubated at 37°C for 15 minutes to fully phosphorylate the primer. Following one round of phenol:chloroform (3:1) extraction, the primer was ethanol precipitated and resuspend to a final concentration of 30 µM in water.

The 10 bp DNA ladder (Invitrogen, 10821–015) was labeled by PNK exchange: 0.4 µg of 10 bp ladder was combined with 50 pmol of ATP and 20 U of T4 PNK in 50 mM imidazole pH 6.4, 12 mM MgCl₂, 1 mM 2-mercaptoethanol, and 70 µM ADP in a total volume of 10 µL and incubated at 37°C for 40 minutes, and then extracted once with phenol:chloroform (3:1), ethanol precipitated, and resuspended in 9 M urea, 100 mM EDTA, 0.2% w/v xylene cyanol, and 0.2% w/v bromophenol blue.

Adapter adenylation

The 3′ DNA adapter was purchased from IDT with a 5′ phosphate and a 3′ inverted dT blocking group:/5Phos/AGATCGGAAGAGCACACGTCTAGTTCTACAGTCCGACGATC/3invdT/. Adenosine 5′-phosphorimidazolide (ImpA) was synthesized according to published protocols [35,36]. The 3′ DNA adapter was 5′-adenylated in 50 mM freshly dissolved ImpA and 25 mM MgCl₂ at 50°C for 3 hours, and then purified on a gel containing 20% acrylamide:bisacrylamide (29:1), 7 M urea, 89 mM Tris–borate (pH 8.3), and 2 mM Na₂EDTA by the crush-and-soak method.

Reverse transcription of RNA oligos

For annealing, 1 pmol of Primer A was mixed with 0.8–1 pmol of CMC-modified or unmodified Oligo Ψa in 50 mM Tris-Cl (pH 7.4) and 100 mM KCl in a total volume of 2.5–5 µL. The primer–template mixture was incubated at 93°C for 2 minutes, and then on ice for 4 minutes. Reverse transcription with either 0.6 U/µL AMV RT (New England BioLabs, M0277T) or 0.28 U/µL HIV RT (Worthington Biochemical Corporation, LS05003) was conducted in 50 mM Tris–acetate (pH 8.3), 75 mM potassium acetate, 8 mM magnesium acetate, 10 mM DTT, and 0.5 mM of each dNTP in a total volume of 5–10 µL. Reverse transcription with SuperScript III RT (Thermo Scientific, 18080044) was conducted in 50 mM Tris–Cl (pH 8.3), 75 mM KCl, 10 mM DTT, and the specified concentration of MgCl₂ or MnCl₂ in a total volume of 5–10 µL. The reactions were incubated at 42°C for 15 minutes for AMV RT, at 37°C for 1 hour for HIV RT, or at 42°C for 3 hours for SuperScript III RT. Next, 10 µL of 9 M urea, 100 mM EDTA, 0.2% w/v xylene cyanol, and 0.2% w/v bromophenol blue were added, and the samples were incubated at 90°C for 5–10 minutes before loading on a gel containing 20% acrylamide:bisacrylamide (29:1), 7 M urea, 89 mM Tris–borate (pH 8.3), and 2 mM Na₂EDTA. After gel electrophoresis, primer extension was visualized by phosphorimaging.

Preparation of sequencing libraries

Human embryonic kidney (HEK) cell line HEK293T/17 was obtained from the American Type Culture Collection (ATCC) and cultured in Dulbecco's Modified Eagle's Medium (DMEM) with high glucose and L-glutamine, without sodium pyruvate (HyClone, SH30022.01) in a 37°C incubator with a humidified atmosphere of 5% CO₂. Total RNA was isolated from confluent 100-mm plates of HEK293T cells by Trizol extraction (Life Technologies, 15596–018) and size-selected for RNAs over 200 nucleotides long with the mirVana miRNA isolation kit (Life Technologies, AM1561). The integrity of size-selected total RNA was checked on a gel containing 1.2% w/v agarose, 89 mM Tris–borate (pH 8.3), 2 mM Na₂EDTA, and 0.4 µg/mL ethidium bromide. The RNA was diluted to 125 ng/µL and fragmented in 50 mM Tris–Cl (pH 7.9) and 8 mM MgCl₂ at 94°C for 10 min, and then purified with an Oligo Clean & Concentrator column (Zymo Research), yielding a total of 34 µg of fragmented RNA. The size distribution of fragmented RNA was found to be centered around 200 nucleotides on a gel containing 1.5% w/v agarose, 89 mM Tris–borate (pH 8.3), 2 mM Na₂EDTA, and 0.4 µg/mL ethidium bromide. The 50%-Ψ standard was added at 0.06 pmol per 1 µg of fragmented RNA, corresponding to 0.03 pmol/µg each of Oligos Ψa and Ua. The RNA was separated into three parts with 11.3 µg RNA each: two parts for CMC treatment (CMC+) and one part for mock treatment (CMC−). The RNA in all three parts was denatured in a total volume of 10 µL at 80°C for 2 minutes, and then placed on ice. The denatured RNA was combined with 20 µL of 50 mM Tris–Cl (pH 8.3), 4 mM EDTA, 7 M urea (TEU buffer) and either 20 µL of TEU buffer (for CMC−) or 20 µL of 1 M CMC freshly dissolved in TEU buffer for a final concentration of 0.4 M CMC (for CMC+). The reactions were incubated at 40°C for 30 minutes, after which the RNA was purified with an Oligo Clean & Concentrator column (Zymo Research) and eluted with 20 µL of water. Next, 40 µL of 50 mM sodium carbonate and 2 mM EDTA (pH 10.4) were added, and the reaction was incubated at 45°C for 2 hours. The samples were ethanol precipitated and resuspended in 20 µL of water.

The RNA was combined with 1 U of calf intestinal alkaline phosphatase (New England BioLabs, M0290S) and 40 U of T4 PNK in 1 × PNK buffer in a total volume of 30 µL, and the dephosphorylation reaction was incubated at 37°C for 30 minutes. After extraction with phenol:chloroform (3:1), the RNA was ethanol precipitated and resuspended in 20 µL water. The 5′-adenylated 3′ DNA adapter was ligated to the repaired RNA at an estimated 1:2 molar ratio of adaptor to RNA (24.25 pmol adaptor for ∼48.5 pmol or 3.3 µg of 200-nucleotide-long fragmented RNA) with 10 U/µL of T4 RNA ligase 2, truncated KQ (New England BioLabs) in 50 mM Tris–Cl (pH 7.5), 10 mM MgCl₂, 1 mM DTT, 15% v/v PEG8000 in a total volume of 40 µL at 16°C overnight (∼16 hours). The RNA was purified with an Oligo Clean & Concentrator column (Zymo Research) and eluted in 14 µL of water. The 5′ ³²P-labeled Primer B (36.3 pmol each for CMC+ and CMC−) was annealed to the RNA in 50 mM Tris-Cl (pH 7.4) and 100 mM KCl in a total volume of 16.5 µL at 93°C for 2 minutes, and then placed on ice for 3 minutes. The CMC+ and CMC− RNA samples were each split into three parts for reverse transcription, where each part contained 1 µg RNA and 11 pmol 5′ ³²P-labeled Primer B. Reverse transcription was conducted in 10-µL reactions, with (1) HIV RT: 0.28 U/µL HIV RT, 50 mM Tris–Cl (pH 8.3), 75 mM potassium acetate, 8 mM magnesium acetate, 10 mM DTT, and 0.5 mM of each dNTP at 37°C for 1 hour; (2) SIII RT, Mn²⁺: 10 U/µL SuperScript III RT, 50 mM Tris–Cl (pH 8.3), 75 mM KCl, 3 mM MnCl₂, 10 mM DTT, and 0.5 mM of each dNTP at 42°C for 3 hours; or (3) SIII RT, Mg²⁺: 10 U/µL Superscript III RT, 50 mM Tris–Cl (pH 8.3), 75 mM KCl, 8 mM MgCl₂, 10 mM DTT, and 0.5 mM of each dNTP at 42°C for 3 hours. To end the reverse transcription reaction, 10 µL of 9 M urea, 100 mM EDTA, 0.2% w/v xylene cyanol, and 0.2% w/v bromophenol blue were added, and sample was incubated at 90°C for 5 minutes. The cDNA was separated on a gel containing 7.5% acrylamide:bisacrylamide (29:1), 7 M urea, 89 mM Tris–borate (pH 8.3), and 2 mM Na₂EDTA, along with 5′³²P-labeled 10 bp DNA ladder (Invitrogen, 10821–015) and 5′ ³²P-labeled Primer B as size controls. After gel electrophoresis, primer extension was visualized by phosphorimaging. The 55-to-260-nucleotide region of each lane was excised, and the cDNA was eluted in 50 mM potassium acetate and 200 mM KCl (pH 7.5) by the crush-and-soak method, and then ethanol precipitated and resuspended in water.

The cDNA was circularized with 5 U/µL of CircLigase II (Epicentre, CL4115K) in 33 µM Tris–acetate (pH 7.5), 66 µM potassium acetate, 0.5 mM DTT, 2.5 mM MnCl₂, and 1 M betaine at 60°C overnight, and then the ligase was inactivated at 80°C for 10 minutes. The cDNA was extracted with phenol-chloroform (3:1) and with chloroform, and then ethanol precipitated and resuspended in water. The cDNA was amplified with 200 µM Illumina multiplex primer and 200 µM Barcoded primer in 1 × Phusion High-Fidelity PCR Master Mix with HF buffer (Thermo Scientific, F531) in a total volume of 50 µL: 30 seconds at 98°C; 12x: 10 seconds at 98°C, 30 seconds at 60°C, 30 seconds at 72°C; 5 minutes at 72°C; hold at 4°C. The cDNA was extracted with phenol:chloroform (3:1) and with chloroform, and then ethanol precipitated. The cDNA was then purified on a 6% TBE minigel (Novex, EC6265BOX) with a 100-base-pair DNA ladder with ethidium bromide staining by the crush-and-soak method, ethanol precipitated, and resuspended in 15 µL of water. The library was checked by Bioanalyzer capillary electrophoresis and by quantitative PCR, and then sequenced in a single lane by Illumina HiSeq 4000 with paired-end 100-base-pair reads at the University of Chicago Genomics Facility.

Mapping of sequencing data

Standard quality control using FastQC was performed after the sequencing and trimming steps. The sequencing reads were processed first with Trimmomatic v0.32 and then with custom Python scripts designed to remove any additional artifacts from demultiplexing and removal of primers, adapters, and low-quality sequences [37]. Next, the trimmed sequences were simultaneously aligned to the 18S and 28S rRNA reference sequences (NR_003286 and NR_003287.2) using Bowtie 1.0 with the highest allowed mismatch settings, yielding 20–50 million 18S and 28S rRNA aligned reads per sample.

The sequence alignment/map (SAM) output from Bowtie was further processed using Python scripts, first to remove a single nucleotide from either end of each read (end-trimming), and then to determine the total read count (c), the number of mutations (m), and the number of stops (s) at each position of the 18S and 28S rRNA reference sequences [37]. Stops at nucleotide position n in the rRNA sequence correspond to reads with a 5′-most nucleotide aligning to the n+2 position, which accounts for end-trimming. The stop rate at each position in the human 18S and 28S rRNAs was calculated as s / (c + s). The mutation rate at each position in the human 18S and 28S rRNAs was calculated as m / (c + s). The mutation rate without counting stops was calculated as m / c. For total mutation rate, m was the sum of A, C, and G reads; for the mutation rate to A, C, or G, m was the number of A, C, or G reads. The mismatch index (MI) at each position was calculated as stop rate + mutation rate. The Δstop and Δmutation rates, and Δmismatch index, were calculated as: rate or index in the CMC-treated sample (CMC+) − rate or index in mock-treated sample (CMC−).

To determine deletion rates, the trimmed sequences were aligned to the 18S and 28S rRNA reference sequences using Bowtie 2.0, and then visualized using the Integrative Genomics Viewer. At each position in the reference sequences, the number of deletions (d) and the total read count (c) at that position were used to calculate the deletion rate as d / c.

Identification of pseudouridine sites

A list of modified sites in human 18S and 28S rRNAs was obtained from the Fournier lab's 3D rRNA modification maps database [30], and the nucleotide positions were re-assigned according to the National Center for Biotechnology Information sequences for human 18S rRNA (NR_003286) and 28S rRNA (NR_003287.2). Having excluded 2′-O-methylated pseudouridine, 2′-O-methylated uridine, and 1-methyl-3-(3-amino-3-carboxypropyl)pseudouridine from the analysis, we counted 92 known pseudouridines and 1047 unmodified uridines in the human 18S and 28S rRNAs.

The context-dependent thresholds for pseudouridine identification were chosen by separately evaluating possible thresholds for each possible +1 nucleotide context. For +1 A, Δstop rate and ΔMI thresholds were evaluated. For +1 C and +1 G, combinations of ΔMI and Δmutation rate thresholds were evaluated. For +1 U, combinations of Δstop rate, Δmutation rate, and ΔMI thresholds were evaluated. In each case, receiver operating characteristic (ROC) curves were plotted, and the threshold with the highest performance in sensitivity, specificity, positive predictive value, and negative predictive value was chosen. The optimized thresholds for all four +1 nucleotide contexts were combined to define the context-dependent threshold shown in Fig. 3A–B (+1 A: Δstop > 0.2; +1 C: ΔMI > 0.15 OR Δmutation > 0.03; +1 U: Δstop > 0.2 OR Δmutation > 0.04; +1 G: ΔMI > 0.09 AND Δmutation > 0.03).

The ROC curve for ΔMI threshold was formed by varying the ΔMI threshold from 0.1 to 0.45; the ROC curve for Δstop threshold was formed by varying the Δstop rate threshold from 0.08 to 0.4; the ROC curve for Δmutation threshold was formed by varying the Δmutation rate threshold from 0.02 to 0.12. The ROC curve for context-dependent thresholds was formed by simultaneously varying the ΔMI, Δstop rate, and/or Δmutation rate thresholds for all four +1 nucleotide contexts: for +1 A, the Δstop rate threshold was varied from 0.08 to 0.4; for +1 C, the ΔMI threshold was varied from 0.15 to 0.35, and the Δmutation rate threshold from 0.025 to 0.03; for +1 G, the ΔMI threshold was varied from 0.07 to 0.15, and the Δmutation rate threshold from 0.012 to 0.08; for +1 U, the Δstop rate threshold was varied from 0.2 to 0.25, and the Δmutation rate threshold from 0.015 to 0.04. Context-dependent thresholds within these ranges were compared, and the thresholds with the high sensitivity and specificity were used to form the ROC curve shown in Fig. 3D.

Data deposition

The sequencing data have been deposited to National Center for Biotechnology Information Gene Expression Omnibus database under accession number GSE110247.

Funding Statement

This work was supported by the National Institutes of Health (RM1HG008935 to T.P., F30GM120917 to K.I.Z., and K01HG006699 to Q.D.), the National Institutes of Health Medical Scientist Training Program grant (T32GM007281), and the University of Chicago Biological Sciences Division and Frank Family Endowment (K.I.Z.). The University of Chicago Genomics Facility is supported by the Cancer Center Support Grant (P30CA014599).

Disclosure of potential conflicts of interest

No potential conflicts of interest were disclosed.

Acknowledgment

We would like to acknowledge past and present members of the Pan laboratory for discussion, as well as the University of Chicago Genomics Facility for high-throughput sequencing.

Financial disclosures

None.