Synthesis of modified nucleotide polymers by the poly(U) polymerase Cid1: application to direct RNA sequencing on nanopores

Abstract

Understanding transcriptomes requires documenting the structures, modifications, and abundances of RNAs as well as their proximity to other molecules. The methods that make this possible depend critically on enzymes (including mutant derivatives) that act on nucleic acids for capturing and sequencing RNA. We tested two 3′ nucleotidyl transferases, Saccharomyces cerevisiae poly(A) polymerase and Schizosaccharomyces pombe Cid1, for the ability to add base and sugar modified rNTPs to free RNA 3′ ends, eventually focusing on Cid1. Although unable to polymerize ΨTP or 1meΨTP, Cid1 can use 5meUTP and 4thioUTP. Surprisingly, Cid1 can use inosine triphosphate to add poly(I) to the 3′ ends of a wide variety of RNA molecules. Most poly(A) mRNAs efficiently acquire a uniform tract of about 50 inosine residues from Cid1, whereas non-poly(A) RNAs acquire longer, more heterogeneous tails. Here we test these activities for use in direct RNA sequencing on nanopores, and find that Cid1-mediated poly(I)-tailing permits detection and quantification of both mRNAs and non-poly(A) RNAs simultaneously, as well as enabling the analysis of nascent RNAs associated with RNA polymerase II. Poly(I) produces a different current trace than poly(A), enabling recognition of native RNA 3′ end sequence lost by in vitro poly(A) addition. Addition of poly(I) by Cid1 offers a broadly useful alternative to poly(A) capture for direct RNA sequencing on nanopores.

INTRODUCTION

RNA plays dual roles during gene expression, both carrying information and interpreting it along the way. RNAs containing protein coding information from genes are acted upon by noncoding RNAs during mRNA processing, translation, and decay. Correct reformulation and use of primary gene transcripts in context critically depend on intricate enzymes with noncoding RNA subunits: the spliceosome, ribosome, and miRNA complexes among others. The composition of cellular RNA populations, both coding and noncoding, reflect cell state. Tracking these states over time can reveal what a cell has done, what it is doing at the moment, and what it might soon do. Rapidly evolving methods for high-throughput RNA sequencing have diversified and multiplied to the extent that our ability to detect and count RNAs by their structure, modification status, location, and associations with other cell components is limited only by the RNA populations we can capture (Pachter 2013). In turn the development of these technologies has relied on foundational research into the capabilities of proteins that manage nucleic acid transactions, and the application of these enzymes in innovative ways.

One such useful enzymatic activity is template-independent nucleotide addition to the 3′ ends of RNA and DNA. The discovery and use of polynucleotide phosphorylase to create RNA polymers can be traced to the origins of molecular biology (Grunberg-Manago et al. 1956; Grunberg-Manago 1989). Terminal deoxynucleotidyl transferase played a key role in early cloning strategies (Jackson et al. 1972), and more recently applications of poly(A) polymerases for labeling or tailing of RNA have been developed (Winter and Brownlee 1978; Lingner and Keller 1993; Martin and Keller 1998). As RNA 3′ nucleotidyl transferases have been studied more broadly, structurally related enzymes that add uridine as well as other nucleotides have been characterized (Preston et al. 2019; Shukla et al. 2020; Liudkovska and Dziembowski 2021), suggesting that a variety of homopolymeric or mixed polymer 3′ tails other than poly(A) could be created for different purposes, including barcoding. Although many studies have tested for 3′ nucleotidyl transferase activity using the four standard ribonucleotides, few studies have addressed the extent to which modified or unnatural bases can be used as substrates. In general, understanding the ability of 3′ nucleotidyl transferases to create nonnative homogeneous or mixed polymer 3′ tails could lead to technological applications that might enable further discovery.

A technology arena that awaits improvements is direct RNA sequencing using nanopores, implemented commercially by Oxford Nanopore Technologies, Inc. (Garalde et al. 2018; Workman et al. 2019). Advantages of direct RNA sequencing by nanopores include the ability to acquire information from full-length single molecules that is lost during the fragmentation necessary for short-read sequencing. Perhaps more intriguing is the potential to read modified nucleotides (Leger et al. 2019; Smith et al. 2019; Begik et al. 2021), whose presence is usually erased by reverse transcription in DNA-based RNA sequencing methods. Despite these advantages, direct RNA sequencing on nanopores suffers from low throughput and limited ability to pool samples within the same sequencing run. Although custom adapters can allow individual RNAs to be targeted (e.g., Smith et al. 2019), the commercial method targets poly(A)+ mRNAs, leaving most nonpolyadenlyated noncoding RNAs (ncRNAs) inaccessible (Garalde et al. 2018; Workman et al. 2019).

Our interest in accessing sequence and modification status for more of the transcriptome led us to propose that addition of unnatural or rare nucleotides to the 3′ ends of RNA could both allow broad capture of more diverse populations of RNA without loss of native 3′ end information. Furthermore, distinct homopolymers or mixed polymers offer the potential for barcoding and pooled sequencing of multiple samples on nanopores. To this end we explored the ability of available 3′ nucleotidyl transferases to incorporate nucleotides other than the standard four that make up most transcripts in the cell. Here we document several novel in vitro activities of the Cid1 poly(U) polymerase (Rissland et al. 2007), including the ability to add long inosine tails to the 3′ ends of a wide variety of RNAs. For unknown reasons Cid1 addition of inosine to polyadenylated mRNAs stalls after addition of 50 residues. Addition of modified U residues is also unusual: 5methylUTP and 4thioUTP are readily incorporated, but ΨTP and 1meΨTP are not. Finally we demonstrate the application of the novel inosine activity in direct RNA sequencing on nanopores, enabling measurement of both mature and nascent mRNA populations as well as noncoding RNAs such as telomerase RNA, spliceosomal snRNAs, snoRNAs, RNaseP RNA, and others.

RESULTS

Cid1 poly(U) polymerase can efficiently add inosine to the 3′ ends of RNA

To find new ways to add modified nucleotides to the 3′ ends of RNA, we tested commercial preparations of two well-studied enzymes, the S. pombe poly(U) polymerase Cid1 (Rissland et al. 2007), and S. cerevisiae poly(A) polymerase (PAP) (Martin and Keller 1998). We first confirmed that the commercial preparations had the expected nucleotide adding specificities by incubating them with various rNTPs and a 24-nt oligomer of adenosine (A24) under the reported conditions (Fig. 1A,B). Briefly, Cid1 from New England Biolabs efficiently adds U and A, but only poorly adds C or G (Fig. 1A, compare lanes 2 and 5 with lanes 3 and 4), in agreement with previous work (Rissland et al. 2007). PAP from Thermo Fisher efficiently adds A, and to a much lesser extent G, but only poorly adds U or C (Fig. 1B, compare lanes 2 and 4 with lanes 3 and 5), as observed previously (Martin and Keller 1998). We next tested whether the enzymes could use inosine triphosphate (ITP, a purine similar to G) to make poly(I) tails. Cid1 added long stretches of inosine to A24 (Fig. 1A, lane 6) whereas PAP only added a few residues (Fig. 1B, lane 6). Addition of inosine by Cid1 was unexpected since the enzyme does not use GTP very well (Fig. 1A, lane 4, Rissland et al. 2007), and inosine differs only by the lack of the 2-amino group present on guanosine.

FIGURE 1.

Nucleotide-adding activity of S. pombe Cid1 and S. cerevisiae Poly(A) polymerase (PAP) on various model RNA substrates. RNA was incubated with enzyme and rNTPs as indicated and run on denaturing polyacrylamide gels, stained with SYBR Gold and imaged on a Typhoon scanner. Markers are DNA with the indicated chain lengths for the equivalently migrating RNA, using the ∼1.04× greater mass to charge ratio of RNA per residue. For example, a 500 nt DNA marks the approximate migration of a 482 residue RNA. (A) Cid1 adds long tails of A, U, or I, but much shorter tails of C or G to A24. (B) PAP adds long tails of A, a 35–50 nt stretch of G, but only short stretches of C, U, or I to A24. (C) Cid1 efficiently adds long stretches of A or U, but mostly just 50 residues of I to a model poly(A)+ mRNA, whereas it inefficiently adds A, U, or I to a nonpolyadenylated RNA. (D) PAP efficiently adds A to either polyadenylated or nonpolyadenylated RNA (lanes 3 and 6) and adds a short stretch of I to poly(A)+ RNA, but only inefficiently adds U or I to nonpolyadenylated RNA.

To test substrates more representative of natural mRNAs, we prepared model mRNAs based on human MYL6 with or without a 40 nt poly(A) tail, to represent polyadenylated [MYL6(A+)] and nonpolyadenylated [MYL6(A−)] RNAs. As with the A24 substrate, Cid1 used UTP or ATP to generate long poly(U) or poly(A) tails on both model RNAs (Fig. 1C, lanes 3, 4, 6, 7), see also (Rissland et al. 2007; Lunde et al. 2012; Munoz-Tello et al. 2012; Yates et al. 2012). Unexpectedly, Cid1 uniformly added about 50 inosine residues to a majority of MYL6(A+) molecules, but also produced a few molecules with long (>200 nt) tails (Fig. 1C, lane 5). In contrast, Cid1 inefficiently added long tails on a fraction of MYL6(A−) molecules (Fig. 1C, lane 8). The low efficiency of Cid1 addition of U, A, or I to this molecule (Fig. 1C, compare lane 8 with lanes 6 and 7) suggests poor access to the 3′ end of the substrate RNA rather than a preference for a particular rNTP substrate.

In comparison, PAP efficiently added poly(A) to both MYL6(A+) and MYL6(A−) RNAs, but only poorly added U to either (Fig. 1D, compare lanes 3 and 4 to lanes 6 and 7, see also Martin and Keller 1998). The presence of a poly(A) tail on MYL6(A+) allowed PAP to add short heterogenous stretches of inosine to most of the molecules (Fig. 1D, lane 5), but MYL6(A−) is much less efficiently used (Fig. 1D, lane 8). When we tested Cid1 and PAP using a shorter (200 nt) model mRNA with 44 A residues [GLuc200(A+)] or without a poly(A) tail [GLuc200(A−)], we saw comparable activities (Supplemental Fig. S1), except that, similar to its activity on the A24 substrate (Fig. 1A, lane 6), Cid1 inefficiently added long poly(I) tails and did not produce more than a hint of a +50 product using GLuc200(A+) (Supplemental Fig. S1A, lane 5). Overall, both enzymes appear more active on substrates with a poly(A) tail, but other substrate features (size, structural features, 3′ end accessibility) may play a role in recognition by these enzymes. Together the results from the model substrates suggest that formation of the unique +50 inosine product by Cid1 is promoted by the presence of a poly(A) tail of at least 40 nt on an RNA of greater than 200 nt. Otherwise Cid1 produces long tails of inosine with an efficiency that depends on the particular substrate RNA.

Validation of inosine addition by Cid1

To precisely measure the lengths of the poly(I) extension of RNA catalyzed by Cid1, we labeled various I-tailed and untailed control RNAs at their 3′ ends with ³²P-pCp and T4 RNA ligase. We then digested the labeled RNA with RNase A, which cuts only after pyrimidines, leaving homopurine polymers like poly(A) or poly(I) intact and still carrying the radioactive phosphate (Fig. 2A). For example, MYL6(A+) contains a U followed by two Gs at its 3′ end to which the template encoded 40 nt poly(A) tail is added during T7 transcription. Thus, the MYL6(A+) substrate labeled in this way is predicted to have a 43 nt RNase A-resistant poly(A) containing oligomer GGA₄₀*pCp (* indicates the radioactive phosphate, designated as GGA₄₀ in Fig. 2B). In the I-tailed MYL6(A+) digest, the 3′ end product is ∼93 nt, confirming that Cid1 adds ∼50 inosine residues onto the preexisting poly(A) tail. The nonpolyadenylated RNAs MYL6(A−) and yeast 5.8S rRNA acquire a heterogeneous I-tail that can be much longer than 50 residues (Fig. 2C,D). Accordingly, RNase A digestion of each of these ³²P-pCp labeled Cid1 products generates a ladder of RNase A-resistant products that extends far up the gel (Fig. 2C,D). We conclude that Cid1 generally adds a uniform ∼50 nt I-tail to poly(A)+ RNAs, whereas it adds from a few to >1000 inosines to nonpolyadenylated RNAs.

FIGURE 2.

Biochemical determination of inosine tail lengths. (A) Scheme for analyzing inosine tails based on RNase A resistance of purine homopolymers. (Top) Poly(A)+ mRNAs, (bottom) nonpolyadenylated RNAs. (Left to right) Tailing by Cid1, 3′ end labeling by T4 RNA ligase and ³²P-pCp, RNase A digestion. (B) Cid1 adds ∼50 inosine residues onto the poly(A) tail of a model poly(A)+ mRNA. (Left) Labeled products without (−) and with (+) I-tailing, and without RNase A digestion, run on a 6% denaturing polyacrylamide gel. (Right) Labeled products without (−) and with (+) I-tailing after RNase A digestion, run on a 12% denaturing polyacrylamide gel. Markers are indicated as for Figure 1. (C) Cid1 adds a long heterogeneous tract of inosine residues onto the 3′ end of a model nonpolyadenylated mRNA. Lanes are as for B above. (D) Cid1 adds a long heterogeneous tract of inosine residues onto the 3′ end of the nonpolyadenylated 5.8S rRNA. Lanes are as for B above.

To confirm that addition of the ∼50 I-tail is an intrinsic feature of Cid1 and not generated by an unknown step in the commercial preparation of the enzyme at New England Biolabs, we cloned, expressed, and purified a recombinant Cid1 with a truncated amino terminus (vCID1) in E. coli (see Supplemental Methods). We found that the addition of ∼50 inosines to MYL6(A+) is a property of both Cid1 preparations and is not dependent on unknown commercial purification or treatment steps (Supplemental Fig. S2A).

Testing the ability of Cid1 to use modified uridine triphosphates

Given its native uridylation activity, we also tested the ability of Cid1 to incorporate modified U residues (Fig. 3), both alone (at 1 mM) and in combination with unmodified UTP (0.5 mM each). We find that Cid1 is unable to add long tracts of 2′-O-methyl-UTP to RNA (Fig. 3, lane 3), and is blocked in the addition of UTP when 2′-O-methyl-UTP is present in the same reaction (lane 4), suggesting that incorporation of 2′-O-methyl-UTP is chain terminating for the Cid1 reaction. Cid1 is also unable to add ΨTP efficiently, however ΨTP does not greatly inhibit incorporation of UTP (Fig. 3, lanes 5 and 6). The same is true for 1meΨTP (Fig. 3, lanes 11 and 12), indicating that ΨTP and 1meΨTP are inactive substrates and poor competitive inhibitors of UTP incorporation by Cid1. In contrast, Cid1 can use either 5me-UTP or 4thio-UTP, producing long tails with or without the addition of UTP (Fig. 3 lanes 7, 8, 9, 10). We conclude that the ability of Cid1 to add 5me-UTP or 4thio-UTP homopolymers (or mixed modified polymers), but not ΨTP or its derivatives, has potential for additional development of nanopore RNA sequencing methods.

FIGURE 3.

Ability of Cid1 to use modified UTP analogs for tailing. Cid1 tailing reactions were set up with the indicated analog either without (−, 1mM) or with (+, 0.5 mM each) competing unmodified UTP, using the MYL6(A+) model mRNA as a substrate. The 6% denaturing polyacrylamide gel was stained with SYBR Gold and imaged on a Typhoon Imager.

Inosine tails generate a distinct signal in the nanopore during sequencing

The utility of modified homopolymers or mixed homopolymers for direct RNA sequencing depends on whether they create a distinct and recognizable signal in the nanopore. To test this, we used a splint ligation method to append one to four copies of a defined synthetic inosine 15mer (I₁₅) to the 3′ end of GLuc200(A+) and GLuc200(A−) RNAs (see Supplemental Methods, Supplemental Fig. S2B). For library preparation, we used a custom adapter with an oligo(dC) segment of 10 residues ([dC]10 adapter) that would pair with the inosine tail (Fig. 4A). We purified RNA with or without a splint-ligated 30 nt poly(I) homopolymer, and with or without a poly(A) tail, prepared libraries using the appropriate adapter oligonucleotide, then sequenced the libraries on nanopores.

FIGURE 4.

Nanopore ionic current signals for poly(I)-tailed RNA molecules. (A) Schematic for Nanopore adaptation of poly(I)-tailed RNA molecules. This step uses a poly(dC)10 oligomer (oligoC adapter) that anneals to the terminal 10 nt of the added inosine tail. (B) Ionic current traces produced by the GLuc200 model mRNA (IV, blue) with or without a poly(A) tail (III, purple) and with or without a ligated 30 nt segment of inosine (II, green). Adaptor sequence trace is shown in gray (I). RNA molecules enter the pore 3′ end first, so the trace records transit through the pore in the 3′ to 5′ direction. (C) Ionic current traces produced by translocation of native yeast TDH3 mRNA or 25S rRNA, with or without I-tailing by Cid1. Segments are colored as in B.

A raw current trace for single representative molecules from each library is shown in (Fig. 4B). Direct RNA sequencing in the ONT nanopore format threads the 3′ end of the RNA into the pore first, and the current (in picoamperes [pA], y-axis) across the pore is captured over time (x-axis) as the molecule transits the pore in the 3′ to 5′ direction. GLuc200(A−) (with no tail of any kind, but with the sequencing adapter ligated directly to its 3′ end) produces a trace showing the adapter sequence (I, gray) followed immediately by the sequence of the GLuc200 body (IV, blue, top panel). In the second panel, the splint-ligated GLuc200(A−)I₃₀ molecule shows a monotonic signal at about 100 pA representing the 30 inosines (II, green) between the adapter (I, gray), and the complex sequence trace (IV, blue). By comparison, GLuc200(A+) lacking any inosines (third panel from the top) shows a monotonic signal corresponding to poly(A) (III, purple) between the adapter and the complex sequence, also around 100 pA. The bottom panel shows a molecule of GLuc200(A+)I₃₀ in which the inosine tail (II, green) was splint ligated to the 3′ end of the poly(A) tail (III, purple). Although the mean ionic current associated with poly(I) and poly(A) homopolymers is similar, the ionic current noise around this mean is substantially different, with a variance of 15.1 ± 6.55 pA for poly(I) compared to 4.3 ± 3.49 pA for poly(A) (Fig. 4B). In addition, there is a distinctive drop in the current amplitude (vertical arrows, Fig. 4B,C) during the transition from the poly(I) to the poly(A) homopolymeric segments. We conclude that I-tails produce a distinct signal in the pore that can be distinguished from native RNA, and may be useful in distinguishing signals of native sequences from the nucleotides added for library capture. A similar current trace has been reported using I-tailing with poly(A) polymerase (Drexler et al. 2021).

Applying the Cid1 inosine tailing reaction for capture of complex native RNA samples

Although PAP was able to add short heterogenous I-tails to poly(A)+ RNAs, the uniform +50 I-tails added by Cid1 (Fig. 1) seemed better for generating representative libraries for direct RNA sequencing. Neither enzyme seemed optimal for adding inosine distributivity to all poly(A)− RNAs in a sample as a fraction of the input material remains unreacted. Since Cid1 appeared to add I-tails to poly(A)− molecules more readily than PAP (Fig. 1; Supplemental Fig. S1), we chose to focus on Cid1. To optimize the Cid1 reaction, we incubated MYL6(A+) with Cid1 and ITP under different conditions, and measured production of the ∼50 nt I-tailed product (see Materials and Methods). Under optimal conditions, Cid1 reacted immediately after addition to the reaction mixture (“0” time sample comes from adding enzyme to the reaction at 4°C, mixing and within seconds taking an aliquot to EDTA), and was complete by 40 min (Supplemental Fig. S2C). A small volume reaction with 1.2–2.4 pmol of RNA ends and 2 units of Cid1 from NEB is sufficient to convert nearly all input poly(A)+ molecules to the +50 form (Supplemental Fig. S2D). This is an amount of RNA that can be used as input for the commercially available Oxford Nanopore Direct RNA library protocol.

To test the use of Cid1 I-tailing for complex natural RNA samples, we I-tailed poly(A)+ RNA from Saccharomyces cerevisiae, prepared libraries using the oligoC adapter (Fig. 4A), comparing them to standard poly(A)+ libraries made using the oligo(dT) adapter in the ONT kit (See Supplemental Tables S1, S2 for a summary). Among the reads obtained from these libraries are examples of TDH3 mRNA and 25S rRNA (Fig. 4C, note: poly(A) selection does not completely remove rRNA, see below). As observed for the synthetic mRNA, a native yeast TDH3 mRNA from the standard library shows the monotonic signal originating from the native poly(A) tail (top panel, III, purple) followed by the complex trace from the mRNA body (IV, blue). The second panel (Fig. 4C) shows a TDH3 mRNA that was I-tailed, and as for the synthetic mRNA (Fig. 4B, bottom panel), the poly(I) signal (II, green) has the same mean current amplitude but broader variation than the poly(A) signal (III, purple), and shows the characteristic drop in current amplitude (arrow) at the transition from poly(A) to poly(I). The 25S rRNA does not have a native poly(A) tail, however I-tailing allows its capture during library preparation and sequencing. As expected, only the broad variance I-tail current trace is observed (Fig. 4C, bottom panel, II, green), after which follows the complex trace formed by the 25S rRNA sequence. We conclude that I-tailing can be used to capture and analyze both poly(A)+ and poly(A)− RNA molecules from a complex natural sample, and that the variance or width of current variation observed allows poly(I) to be distinguished from poly(A), at least visually in these traces.

Detection and quantification of poly(A)+ mRNAs using I-tailing

The ability of I-tailing to capture both poly(A)+ and poly(A)− RNAs from the same biological sample potentially extends analysis of the transcriptome beyond mRNA (Fig. 4). However, to determine whether poly(I) tailing captures the mRNA population as faithfully as the direct poly(A) method, we compared multiple libraries of different preparations of yeast poly(A)+ RNA using either the commercial poly(A) capture method or the Cid1 I-tailing poly(I) capture method. Reads were mapped to the genome and counted for each gene, and coverage for each gene was compared (Supplemental Fig. S3). When direct poly(A)-capture libraries are compared to I-tailed and poly(I)-capture libraries made using poly(A)+ mRNA, the efficient capture of residual rRNA reads in the I-tailed libraries distorts the parametric similarity between the libraries, but leaves a strong rank order (nonparametric) correlation intact (Supplemental Fig. S3).

To estimate similarity in mRNA expression measurements across methods, we filtered out rRNA reads and adjusted coverage to reads per million minus rRNA (RPM -rRNA, Fig. 5A).

FIGURE 5.

Quantitative evaluation of 5099 yeast mRNAs with and without I-tailing. (A) Strongly correlated abundance measurements for >5000 yeast mRNAs using either standard poly(A)+ capture or I-tailing by Cid1 and capture with the oligoC adapter. (B) Estimation of median tail lengths by Nanopolish of poly(A)+ mRNAs using direct capture of poly(A)+ with the standard oligo(dT) adapter (black, 30–50 nt) or after tailing and capture with the oligoC adapter (blue, 60–80 nt). (C) Inferred length of I-tail added as a function of mRNA abundance. For each of 4965 poly(A)+ mRNAs, the estimated median poly(A) tail length was subtracted from the estimated median poly(A)poly(I) tail length and plotted versus abundance. Across a wide range of expression levels, Nanopolish estimates the added inosine polymer to be ∼27 nt. The version of Nanopolish used here has only been trained to distinguish and measure poly(A) tails (see text).

Comparison of expression of ∼5000 genes shows that the two methods provide highly similar measurements across the yeast poly(A)+ transcriptome (Fig. 5, Pearson's r = 0.95, Spearman's ρ = 0.95). Additional comparisons are shown in Supplemental Figure S3 and indicate that technical replicates using the poly(I)-tailing method may be slightly more noisy than those using the commercial poly(A)-capture method, as expected for a protocol with an additional step (I-tailing). In addition, the detection of contaminating poly(A)− RNAs (not counting rRNAs which are filtered out) by the I-tailing method suggests that broader capture of snRNAs, snoRNAs, and other nonpolyadenylated RNAs may explain some of the differences (see below). Overall, the two methods seem equally fit for the measurement of poly(A)+ mRNA from complex biological samples, but the I-tailing method captures additional RNAs that are not polyadenylated.

Detecting the homopolymer extension on I-tailed poly(A)+ mRNAs using nanopores

Biochemical analysis indicates that a model poly(A)+ mRNA receives ∼50 new residues during I-tailing (Fig. 2B). To determine if Cid1 adds a uniform length I-tail onto different poly(A)+ mRNA molecules in a complex sample, we analyzed poly(I)-tailed total yeast poly(A) mRNA nanopore reads using Nanopolish, a software package for estimating poly(A) tail lengths (version 0.10.2, Loman et al. 2015). Because Nanopolish is not trained to distinguish between poly(I) and poly(A), it calls the linked poly(I)-poly(A) signal as a single homopolymer segment (Fig. 5B, blue). We plotted the mRNAs by coverage (x-axis) and Nanopolish median tail length estimate (y-axis) from a standard poly(A) capture library (black) and a poly(I)-tailed library (blue) that gave comparable numbers of total reads (Fig. 5B; Supplemental Table S3). The majority of the control poly(A) homopolymer length estimates are in the 30–50 nt range as expected for yeast mRNAs (Sachs and Davis 1989), with the more abundant mRNAs having slightly shorter poly(A) tails on average than less abundant mRNAs (Lima et al. 2017). For the poly(I)-tailed poly(A)+ mRNAs, Nanopolish median tail length (per gene) estimates range from 60–80 nt with an average of 70 nt (Fig. 5B). By subtraction of the median poly(A) alone tail estimates, this suggests a poly(I) tract of about 28–30 nt, well below the 50 nt determined by biochemical analysis (Fig. 2). Nanopolish estimation of homopolymer lengths is at least partially dependent on the rate at which the polymer transits the pore, suggesting that perhaps poly(I), or the combined poly(A + I) polymer transits more quickly than poly(A). Future software development may allow computational discrimination of poly(I) segments from poly(A) segments within each homopolymer tail as well as a more accurate estimate of I-tail lengths. The strong agreement in gene expression measurements made with or without I-tailing indicates that Cid1 uniformly adds about the same length I-tail to each poly(A)+ mRNA, with the exception of very short mRNAs as noted above (Fig. 1; Supplemental Fig. S1).

Capture and measurement of nonpolyadenylated RNAs by I-tailing

To examine the ability of Cid1 to I-tail nonpolyadenylated RNAs, we incubated total and rRNA-depleted yeast RNA samples with Cid1 and ITP (Fig. 6). After Cid1 I-tailing, a slight shift up in migration of the 18S and 25S rRNAs is observed (Fig. 6A, arrows). Because purified 5.8S rRNA acquires a heterogenous I-tail that can be quite long (Fig. 2D), 5.8S rRNA no longer appears as a sharp band on the gel (Fig. 6A). There is little if any distinguishable change in 5S rRNA or the tRNAs, suggesting they may fail to bind the enzyme or their 3′ ends may be buried or otherwise blocked from entering the enzyme active site. These results reinforce observations using model RNAs (Fig. 1) that different nonpolyadenylated RNAs may have different Cid1 tailing efficiencies.

FIGURE 6.

Detection of nonpolyadenylated RNAs. (A) I-tailing of total yeast RNA. Yeast RNA treated with (+) or without (−) Cid1 and ITP for I-tailing were run on a 6% denaturing polyacrylamide gel and stained with SYBR Gold. Slight shift up of 18 and 25S rRNAs indicates addition of poly(I). Disappearance of 5.8S rRNA is consistent with its acquisition of a long heterogeneous tail (See Fig. 2). (B) Genome browser view of the rRNA genes of S. cerevisiae showing detection of rRNA by different library preparation methods. Top line is poly(A) selected RNA using the direct oligoT capture method, second line is poly(A) selected and I-tailed RNA captured with the oligoC adapter, third line is total RNA I-tailed and captured with the oligoC adapter, fourth line is Ribominus rRNA depleted total RNA I-tailed and captured with the oligoC adapter. (C) Scatter plot showing robust detection of nonpolyadenylated noncoding RNAs by I-tailing and capture with the oligoC adapter.

To determine the extent to which I-tailing by Cid1 is useful for detecting and measuring non-poly(A) RNA, we sequenced rRNA from libraries made using different RNA sources by both the standard poly(A) capture method and the I-tailing method. Libraries made from either poly(A) selected RNA (rRNA is a contaminant in poly(A) selected RNA), total RNA, or total RNA subjected to rRNA depletion by the RiboMinus kit were sequenced on nanopores and mapped to the S. cerevisiae rDNA locus (Fig. 6B). As expected, the standard ONT poly(A) capture libraries have <0.1% of reads aligning to the rRNA genes (Supplemental Table S2). By comparison, when poly(A)-selected RNA is I-tailed and captured using the oligoC adapter, residual nonpolyadenylated RNAs contaminating the poly(A) selected sample is detected with ∼4%–8% of the reads comprising rRNA (Supplemental Table S2). In total RNA with no depletion step, I-tailed rRNAs accounted for 73% of the reads (Supplemental Table S2), suggesting that sensitivity of detection of other nonpolyadenylated RNAs would be improved by rRNA depletion. RiboMinus treatment successfully depleted 18S and 25S rRNA, but not 5.8S rRNA (Fig. 6B), which still comprised 41% of the reads from this sample (Supplemental Table S2). Since the much shorter 5.8S rRNA takes far less time than 18 or 25S rRNA to go through the pore, Ribominus rRNA depletion allowed detection of a diverse set of nonpolyadenylated RNAs including snoRNAs (Fig. 6C, black circles), the yeast homologs of telomerase RNA, RNaseP RNA, U3 snoRNA, and all the spliceosomal snRNAs except U6 (which has a 3′ phosphate that is not a substrate for I-tailing, Lund and Dahlberg 1992), and others (Fig. 6C, magenta circles). Reads mapping to the intronic snoRNA genomic coordinates (blue circles) are efficiently captured in poly(A)+ samples as part of the pre-mRNA precursors and processing intermediates within which they reside. Our ability to detect noncoding RNAs is likely a product of both their abundance in the sample and in the accessibility of their 3′ ends to the enzyme.

Capturing nascent transcript structure by Cid1 I-tailing of chromatin-associated RNA

There is growing interest in methods that can capture and directly sequence nascent RNAs, enabling RNA modification and processing to be detected cotranscriptionally (Incarnato et al. 2017; Hsiao et al. 2018; Wissink et al. 2019). We investigated the suitability of I-tailiing for detecting nascent RNA transcripts by enriching for nascent RNAs and isolating chromatin from yeast strain CKY2647 (gift from Dr. Craig Kaplan). CKY2647 has a carboxy-terminal AviTag on the RNAPII RPB3 subunit and constitutively expresses the biotin ligase BirA, which biotinylates the AviTag in vivo (Fairhead and Howarth 2015). Chromatin preparations were incubated with streptavidin beads to capture biotin-tagged RNAPII along with nascent transcripts, and then recovered by phenol-chlorofom extraction (Fig. 7A).

FIGURE 7.

Capture and analysis of RNA polymerase II (RNAPII) associated nascent transcripts by Cid1 I-tailing. (A) Protocol for isolating RNAPII-associated transcripts from chromatin (see Materials and Methods). (B) A model for cotranscriptional splicing. Introns bounded by 5′ splice sites (green) and 3′ splice sites (gold) are removed creating exon–exon junctions (blue) in nascent transcripts as RNAPII (magenta) moves down the gene. (C) Genome browser views of nanopore reads mapping to ACT1 (top), RPS4B (middle), and FBA1 (bottom). The 3′ ends of reads from nascent RNAs are distributed along the gene, and splicing usually occurs before RNAPII has moved 100 nt beyond the 3′ splice site.

After purifying the RNA from this sample, we I-tailed, sequenced, and mapped it to the yeast genome, inferring that the location of the 3′ end of the transcript would identify the position of RNAPII on the gene (Fig. 7B; Churchman and Weissman 2011). To analyze transcripts that initiated at the promoter, we filtered for reads whose 5′ ends mapped within 200 bp of the annotated start codon and aligned these to the genome, comparing them to similarly filtered transcripts from the rRNA depleted I-tailed library (RPS4B, FBA1) or the poly(A) selected I-tailed library (ACT1, Fig. 7C). Whereas >90% of the transcripts found in rRNA depleted or poly(A) selected total RNA have 3′ ends that map together near the end of the transcription unit (as judged by short read coverage, blue tracks, Fig. 7C), the 3′ ends of transcripts from the chromatin-associated RNA were distributed along the gene as expected for nascent transcripts. For the intron-containing genes ACT1 and RPS4B, most of the nascent transcripts have had their introns removed while RNAPII is still engaged, consistent with cotranscriptional splicing. In agreement with other work (Carrillo Oesterreich et al. 2016), a substantial fraction of yeast nascent transcripts is spliced before RNAPII moves more than 100 nt from the 3′ splice site (Fig. 7C). We conclude that following preparation of chromatin-associated RNA, I-tailing by Cid1 is suitable for detection and analysis of processing of nascent transcripts by nanopore direct RNA sequencing.

DISCUSSION

We set out to test the ability of 3′ nucleotidyl transferases to incorporate modified nucleotides in order to explore potential applications for nanopore direct RNA sequencing. Polynucleotide phosphorylase, useful for base nonspecific addition of rNDPs to 3′ ends (Grunberg-Manago et al. 1956; Grunberg-Manago 1989) was initially tested and ruled unsuitable due to 3′ trimming by the competing reverse 3′ phosphorolytic activity (for example, see Unciuleac et al. 2021). Upon testing Cid1 poly(U) polymerase, new activities of Cid1 were revealed (Figs. 1–3), in particular the unexpected synthesis of inosine homopolymers (Fig. 2). We exploited this new activity to capture synthetic and natural transcripts from complex samples for nanopore sequencing (Figs. 4–7). We showed that I-tailing by Cid1 does not distort measurement of poly(A)+ mRNAs (Fig. 5), while simultaneously enabling robust detection of essential nonpolyadenylated RNAs such as snRNAs, telomerase RNA, and RNase P RNA (Fig. 6). Composition of RNAs isolated from chromatin, such as partially processed nascent transcripts, can also be analyzed (Fig. 7).

New activities of Cid1

Our examination of the ability of Cid1 to add modified nucleotides to the 3′ ends of RNA was motivated by limitations in the standard nanopore direct RNA sequencing protocol. The standard protocol targets the natural poly(A) tail present on most mRNAs (Garalde et al. 2018); however, many target RNAs, such as nascent RNAs, mature snRNAs, snoRNAs, rRNAs, and many other noncoding RNAs, lack a poly(A) tail. Enzymatic addition of poly(A) to capture nonpolyadenylated RNAs (Wongsurawat et al. 2019; Drexler et al. 2021) adds information that then cannot be disambiguated from native RNAs undergoing decay or other post-transcriptional events that involve addition of short poly(A) tracts (Tudek et al. 2018), or other 3′-nucleotidylation events (Zigáčková and Vaňáčová 2018; Liudkovska and Dziembowski 2021). ATP analog 6-bioATP (N6-[(6-amino)hexyl]-amino-ATP-biotin is a substrate for Cid1 [Moritz and Wahle 2014]), but evidence for other noncanonical nucleotides was scarce, and we were uncertain whether such a large modification would fit through the pore. Our results indicate that addition of modified nucleotides is a feasible alternative to the standard method and that distinct current signatures allow their discrimination from native nucleotides.

The structure and function of Cid1 has been studied in detail (Rissland et al. 2007; Lunde et al. 2012; Munoz-Tello et al. 2012, 2014; Yates et al. 2012, 2015); however, it remains unclear how ITP is recognized by the enzyme. Crystal structures of Cid1 bound to each of the four standard rNTPs (Lunde et al. 2012), show that while U, A, and CTP are bound in the anti orientation, GTP is bound in the syn orientation, possibly because the guanosine 2-amino group fits poorly into the binding pocket otherwise. Inosine lacks the 2-amino group of guanosine, suggesting it may be able to bind in the anti conformation, similar to ATP or UTP (Lunde et al. 2012). In this orientation, the N1 and O6 of inosine might interact with enzyme functional groups that normally interact with N3 and O4 of its preferred substrate UTP (Lunde et al. 2012; Munoz-Tello et al. 2012; Yates et al. 2012). Additional studies will be needed to explain how inosine is recognized by Cid1.

Our survey of UTP analog use also revealed interesting new properties of the Cid1 enzyme. First, 2′-O-methyl-UTP, strongly inhibits UTP incorporation while supporting extension of the RNA substrate by one or a few nucleotides at most (Fig. 3), suggesting it can be incorporated but is chain terminating. Second, the enzyme uses 5meUTP (aka ribothymidine triphosphate) but not ΨTP or 1meΨTP (Fig. 3). This is surprising because the methyl group on the C5 position of U in 5meUTP occupies the same space as the methylated N1 in 1meΨTP (Fig. 3, note that atoms at positions 5 and 1 are flipped during conversion of UTP to ΨTP, Spedaliere et al. 2004). It is hard to imagine how this difference would affect binding or catalysis given the available crystal structures (Lunde et al. 2012; Munoz-Tello et al. 2012, 2014; Yates et al. 2012). It seems unlikely that the structural basis for discrimination against ΨTP analogs lies in the difference between the glycosidic bond (N1 to C1′) in UTP and 5meUTP as compared to the C–C bond (C5 to C1′) joining base to sugar in ΨTP and 1meΨTP, but other explanations are not evident and will require future experiments.

Addition of inosine tracts to poly(A)+ mRNA stalls after addition of about 50 residues (Figs. 1C, 2B, 5B). Although we have not systematically determined them, the requirements for this strong stop appear to include a specific length of poly(A) greater than 24, since we do not observe this stop in the oligo A24 reaction (Fig. 1A). We have not tested a model mRNA with an A-tail of 24 so it is also possible that additional RNA 5′ of the A tract may be required to produce the +50 stop. A 200 nt model mRNA with a 44 nt poly(A) tail was a substrate for A and U addition, but tailed poorly with I, showing low amounts of long I-tails with no accumulation of +50 nt product [GLuc200(A+), Supplemental Fig. S1A], suggesting that features in the mRNA body may contribute to the quality of binding to Cid1. Most yeast mRNAs have >24 A residues and the high degree of quantitative correspondence between I-tailed libraries with the standard poly(A) capture libraries (Fig. 5) indicates that few natural yeast mRNAs behave like the GLuc200(A+) substrate. What creates the strong stop is unknown, however I-A base “wobble” base-pairing could produce an extended duplex or other structure that might prevent further reaction once the I-tract reaches 50 residues.

RNAs that lack poly(A) tails acquire a heterogeneous length I-tail that ranges from a few to many hundreds of residues long (Figs. 1A,C, 2C,D). The efficiency of I-tailing on individual RNAs lacking poly(A) tails is variable, and the product distribution suggests that initially the reaction is distributive and inefficient until some molecules acquire a short I-tail after which the reaction becomes more processive on that class of molecules. An example of an efficient substrate is 5.8S rRNA (most molecules immediately acquire heterogenous tails), whereas 5S rRNA is barely if at all recognized (Fig. 6). The minimum size for initial substrate recognition by Cid1 is at least 15 residues (Rissland et al. 2007), and we can easily detect 5.8S rRNA (∼120 nt) and several snoRNAs that are less than 100 nt. The characteristic tailing efficiency of each individual poly(A)− RNA will influence its sensitivity of detection. We have not examined the effect of the length of the I-tail on capture efficiency during library preparation or sequencing.

Applications and limitations of the current method and potential for future development

We have demonstrated the utility of Cid1 I-tailing for capture, library preparation, and direct RNA sequencing on nanopores. One particular advance from this effort is the ability to detect many important nonpolyadenylated structural RNAs in the transcriptome (Fig. 6). Because the 3′ ends of many of these RNAs are buried in a structure that helps stabilize them, their Cid1 I-tailing efficiencies will be idiosyncratic, and thus their representation in libraries may not match their abundance in the sample. This limitation is present for any method that requires access to the RNA 3′ end, and it means determination of absolute numbers of molecules or quantification of two such RNAs relative to each other within the same sample may not be possible. Nonetheless, the relative change in the same RNA measured in two samples may be determined, provided the treatment does not alter the 3′ end of the target RNA. Thus, the method will capture changes in both the amount of an RNA and in the structure of its 3′ end. The ends of most nascent RNA transcripts are unlikely to be highly structured as they will be distributed along all positions of the transcription unit, and thus representation should be relatively even across the gene (Fig. 7). Thus, this method enables detection of a large and complex class of RNA that are missed by the standard poly(A) capture method. We also show that I-tailing quantitatively recovers poly(A)+ mRNAs equivalently to the standard method, allowing both poly(A)+ and poly(A)− RNAs to be captured in the same sample.

Our poly(I) sequencing libraries typically have 10%–30% the throughput compared to standard ONT poly(A) sequencing (Supplemental Table S2), potentially a concern for a method that is already limited by throughput. One reason could be that the long I-tails on some RNAs sequester the oligoC adapter, preventing its ligation to other molecules, thus reducing the efficiency of ligation of the motor protein adapter, and reducing the depth of the library. Another possibility is that very long poly(I) tails may stall more frequently in the nanopore. During DNA sequencing in nanopores, molecules that are ejected are unlikely to be sequenced since the motor protein has already migrated across the molecule (Loose et al. 2016). Further development to control the length of the I-tails may address both of these possible limitations to throughput.

I-tailing provides a distinct electronic signature (Fig. 4), potentially allowing RNAs from an I-tailed sample to be distinguished from non-I-tailed molecules in a sequencing experiment where two different libraries have been mixed. Currently the ability to discriminate I-tails from A-tails from I + A-tails is insufficiently accurate to deconvolute read sample origin (L. Mulroney, unpubl.; see also Drexler et al. 2021). In addition, several modified UTP analogs (Fig. 3), as well as at least one N6 modified ATP analog (Moritz and Wahle 2014) can be incorporated by Cid1. No doubt Cid1 can create mixed polymers from combinations of base modified precursors, and these may very well have distinct electronic signatures that will allow a tailing step to double as a barcoding step for direct RNA sequencing in nanopores. Additional 3′-nucleotidyl transferases besides Cid1 may have additional capabilities that have yet to be exploited (e.g., Preston et al. 2019).

MATERIALS AND METHODS

In vitro transcription template generation

DNA templates for MYL6(A+) and MYL6(A−) transcripts were generated using linearized pUC13-MYL6 plasmid. pUC13-MYL6 was digested with BbsI [NEB: R0539S, cleaves after the poly(A) sequence in the template], or BsmI [NEB: R0134S, cleaves before the poly(A) sequence in the template] for MYL6(A+) or MYL6(A−), respectively (see Supplemental Methods for sequence of the MYL6 template). GLuc200(A−) and GLuc200(A+) templates were generated using PCR amplification of GLuc of the first 200 nt at the 5′end of pCMV-GLuc 2 Control Plasmid (NEB: https://www.neb.com/tools-and-resources/interactive-tools/dna-sequences-and-maps-tool). The sequence was targeted using a forward primer (5′-TCGAAATTAATACGACTCACTATAGGGAGACCCAA) containing a T7 promotor region, and a reverse primer that terminates the PCR product at the 3′ end of the truncated GLuc sequence (5′-GCGGCAGCTTCTTGCC) or templates the addition of a 40 nt 3′ poly(A) tail for GLuc200(A−) and GLuc200(A+) (5′-TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTGCGGCAGCTTCTTGCC), respectively. Amplification was obtained using Platinum Taq DNA Polymerase High Fidelity (Invitrogen) in reactions with 1× High Fidelity PCR buffer, 0.01 µg/µL of plasmid, 0.4 mM dNTP Mix, 0.4 µM Forward Primer, 0.4 µM Reverse primer, 2 mM MgSO₄, 1 unit Platinum Taq DNA Polymerase High Fidelity. PCR cycles were: 94°C 30 sec (94°C 10 sec, 58°C 15 sec, 65°C 45 sec) 22 cycles, then 65°C 10 m, 4°C hold. DNA In vitro transcription templates were purified using NucleoSpin Gel and PCR Clean-up (Macherey-Nagel) following the manufacturer's instructions.

T7 in vitro transcription

Templates (described above) were transcribed using the MEGAscript T7 Transcription Kit (Invitrogen). Reaction products were separated using 8 M urea, 6% polyacrylamide gel electrophoresis, identified by staining with ethidium bromide, and excised. Gel slices were rotated at 4°C overnight in RNA Elution Buffer (0.3 M NaOAc pH 5.2, 0.2% SDS, 1 mM EDTA, 10 µg/mL proteinase K). Eluted product was purified using 25:24:1 phenol:chloroform:isoamyl alcohol and ethanol precipitation.

RNA extraction and initial sample preparation

Total RNA from S. cerevisiae BY4741 cells was extracted by a hot phenol method as described previously (Ares 2012). Poly(A) RNA was selected using NEXTflex Poly(A) beads (BIOO Scientific Cat#NOVA-512980) according to the manufacturer's instructions. Ribosomal RNA depletion of total RNA used RiboMinus Transcriptome kits and Concentration Modules (Invitrogen #K155001, K155003) and was done according to the manufacturer's instructions.

Polynucleotide tailing with S. pombe Cid1 Poly(U) polymerase

To test for tailing ³²P-labeled A24 as a substrate using residues (ATP, CTP, GTP, UTP, or ITP) a reaction containing labeled A24 RNA, 0.5 mM NTP, 50 mM NaCl, 10 mM MgCl₂, 10 mM Tris-Cl pH 7.9, 1 mM DTT, 500 µg/mL Bovine Serum Albumin (BSA), and 0.6 units of NEB poly(U) polymerase in a final volume of 5 µL was incubated at 37°C for 30 min. The standard reaction for adding homopolymers to the 3′ends of model mRNAs and yeast RNA preparations using Cid1 poly(U) polymerase is as follows: RNA (various amounts) in 0.1 mM EDTA in volumes up to 2.95 µL is denatured at 95°C for 2 min then placed on ice for 2 min. The RNA is added to a reaction containing 4mM NTP (ATP, CTP, GTP, UTP, or ITP) 50 mM NaCl, 13.5 mM MgCl₂, 10 mM Tris-Cl pH 7.9, 1 mM DTT, 500 µg/mL BSA, and 0.6 units of NEB poly(U) polymerase in a final volume of 7.5 µL and incubated at 37°C for 1 h. To test for incorporation of modified U residues, 200 fmol MYL6(A+) RNA was incubated with 1 mM concentration of modified UTP analog or a combination of 0.5 mM modified UTP analog and 0.5 mM UTP following the standard poly(U) polymerase reaction. Inosine triphosphate was purchased from Sigma. Modified UTP analogs were purchased from TriLink, Inc.

Polynucleotide tailing with S. cerevisiae Poly(A) polymerase

To add a homopolymer to the 3′ ends of RNAs using Poly(A) polymerase, yeast (Thermo Scientific 74225Y/Z) a reaction containing [1× Poly(A) polymerase reaction buffer, 200 fmol RNA, 0.5 mM NTP] is incubated at 37°C for 30 min, then two volumes of Gel Loading Buffer II (Invitrogen: AM8546G) are immediately added to stop the reaction. Reaction products were separated on 15%, 8%, or 6% denaturing polyacrylamide gels depending on their sizes.

pCp labeling

T7 transcripts and their I-tailed counterparts were labeled with pCp [5′-32P] Cytidine 3′, 5′ bis(phosphate) 3000 Ci/mmol, 10 mCi/mL, using NEB T4 RNA Ligase 1, 1× reaction buffer, 0.15 mM ATP, 10% DMSO, 20 µCi ³²P-pCp, 6 pmol of RNA, and 333 units RNA Ligase 1 (NEB M0204S) with incubation at 16°C for ∼16–18 h. The products were purified by extracting with an equal volume of 25:24:1 phenol:chloroform:isoamyl alcohol, and 0.3 mM NaOAc pH 5.2 was added before ethanol precipitation.

RNase A digestion

RNA samples in 100 mM NaCl, 10 mM EDTA, 0.025 µg/µL RNase A (Thermo Scientific RNase A, cat no. EN0531) were incubated at 37°C for 15 min, then immediately vortexed with equal amounts of 25:24:1 phenol:chloroform:isoamyl alcohol. The aqueous phase was made 0.3 M NaOAc pH 5.2 precipitated with ethanol, rinsed with 70% ethanol, briefly dried and then suspended in formamide and dyes for electrophoresis.

Preparation of RNAPII-associated RNA

Growth of yeast

Saccharomyces cerevisiae CKY2647 (MATa ura3-52::BirA::kanMX his3Δ200 leu2Δ met15Δ0 trp1Δ63 lys2-128δ gal10Δ56 rpb1d::CLONAT rbp3::AVITAG-TAP::KITRP1, pRP112 RPB1 CEN URA3) is grown to an A600 of 0.5–0.8 in YEPD in 100 mL cultures. Cells are harvested at 1100g centrifugation for 5 min at 4°C. Pellets are washed with 40 mL ice-cold PBS twice, then transferred to 1.7 mL Eppendorf tubes and washed with cold PBS by centrifugation at 1100g for 5 min at 4°C. Supernatant is removed and pellets are snap frozen in liquid nitrogen before storing in −80°C.

Preparation of chromatin

A hole is pierced in the bottom of a 15 mL Falcon tube with a 22-gauge needle and placed inside a 50 mL Falcon tube cap into which a hole has been cut just large enough to fit the 15 mL tube. A strip of parafilm is wrapped around the 12 mL mark on the 15 mL Falcon tube for stabilization in the 50 mL Falcon tube. The 50 mL Falcon tube cap containing the 15 mL centrifuge is then screwed on to the 50 mL Falcon tube—this assemblage is a make-shift bead filter. Working in a 4°C room, the yeast pellets are retrieved from the −80°C freezer and placed on ice for 5 min, then resuspended in 1 mL of Buffer 1 (20 mM HEPES pH 8.0, 60 mM KCl, 15 mM NaCl, 5 mM MgCl₂, 1 mM CaCl₂, 0.8% Triton-X100, 0.25 mM sucrose, with freshly added 0.5 mM spermine and 2.5 mM spermidine). The cells are lysed in a 2 mL Eppendorf tube containing 1 mL 0.5 mm zirconia beads by six cycles of 1 min vortexing with 1 min pauses using the Turbomix attachment on a Vortex Genie 2 (Scientific Industries Inc. SKU: SI-0564) that had been prerun at max speed for 1 min preceding the vortexing to ensure consistent machine performance in the cold. The sample (plus beads) is transferred to the 15 mL Falcon tube. For transfer of remaining lysed cells left in the 2 mL tube, 1 mL of Buffer 1 was added, then the tube was inverted before transferring to the assembled bead column. This process was repeated three more times, for a total of four times altogether. The assembled Falcon tube bead filter was centrifuged at 400g for 6 min at 4°C to remove the lysate from the zirconia beads. Avoiding the pellet, 2 × 750 µL of the supernatant at the bottom of the 50 mL Falcon tube is transferred into each of two 1.7 mL Eppendorf tubes and the pellet is discarded. The samples are centrifuged at 2000g for 15 min at 4°C, the supernatant fluid is removed and the chromatin pellets are resuspended in 800 mL of Buffer 1, the pairs are combined into one 1.7 mL Eppendorf tube and centrifuged again at 2000g for 15 min at 4°C. Using a p1000, the chromatin pellet is aggressively resuspended in 800 µL of Buffer 2 (20 mM HEPES pH 7.6, 450 mM NaCl, 7.5 mM MgCl₂, 20 mM EDTA, 10% glycerol, 1% NP-40, 2 M urea, 0.5 M sucrose, with freshly added 1 mM DTT and 0.2 mM PMSF). The sample is vortexed for 5–10 sec, then incubated on ice for 5 min. After centrifugation at 2000g for 15 min at 4°C, the pellet is again suspended in 800 µL of Buffer 2. The sample is centrifuged at 2000g for 15 min at 4°C, then the pellet is finally suspended in 300 µL of Buffer 3 (HEPES pH 8.0, 60 mM KCl, 15 mM NaCl, 5 mM MgCl₂, 1 mM CaCl₂) before adding to prepared streptavidin beads for RNAPII capture. Streptavidin beads (NEB #S1420S) are prepared by taking 150 µL of the bead suspension and equilibrating them in 700 µL of bead buffer (0.5 M NaCl. 20 mM Tris pH 7.5, 1 mM EDTA), and recovering them on the wall of the tube with a magnetic rack. The bead buffer wash is discarded and the beads are resuspended in 300 µL of the pellet in Buffer 3 from above. The mixture is rotated at 4°C for 2 h, and then the beads are washed three times as follows: Collect on the tube wall on a magnetic rack, remove and discard supernatant, add 500 µL Buffer 3 to the beads, remove from the magnet and suspend beads. For RNA purification, the beads are collected from the final wash and resuspended in 500 µL of RNA Extraction Buffer (0.3 M NaOAc pH 5.3, 1 mM EDTA, 1% SDS), 100 µL acid phenol, and 100 µL chloroform. This is vortexed vigorously and then centrifuged at 14,000g for 5 min. The aqueous phase is combined with 1.2 mL of 100% ethanol in a fresh 1.7 mL Eppendorf tube, mixed well and incubated at −80°C for 1 h or overnight. Precipitated RNA is recovered by centrifugation, the pellet is rinsed with 70% ethanol and briefly air dried. To remove contaminating DNA, the pellet is resuspended in 100 µL of DNAse solution (1× Turbo DNase Buffer, 10 units TURBO DNase [Invitrogen AM2238]) and incubated at 37°C for 30 min. The sample is finally purified using an RNA Clean and Concentrator-5 kit (Zymo R1013) following the manufacturer's instructions, and eluted in 35 µL of RNase-free water.

Library preparation for direct RNA sequencing on nanopores

Purified RNA (500–775 ng) was prepared for library construction as follows: Poly(A) enriched samples captured by their endogenous poly(A) tail using either the ONT SQK-RNA001 kit or its replacement SQK-RNA002 kit following the manufacturer's instructions. I-tailed RNA was captured using the ONT SQK-RNA002 kit except that a custom oligoC adapter was used in place of the RTA (oligoT) adapter provided by the kit. The optional reverse transcription step was done for all libraries in this study except that Superscript IV (Thermo Fisher) was used in place of Superscript III. To assemble the custom oligoC adapter, 100 pmol of top oligo (5′-pGGCTTCTTCTTGCTCTTAGGTAGTAGGTTC-3′) and 100 pmol bottom oligo (5′-CCTAAGAGCAAGAAGAAGCCCCCCCCCCCC-3′) are mixed in 10 µL of 50 mM Tris pH 8, 100 mM NaCl, 0.1 mM EDTA and incubated in a thermocycler at 75°C, then slow cooled at a ramp rate of 0.1°C/sec to 23°C. The annealed duplex adapter is diluted to 100 µL with 90 µL water, and 1 µL of this 1 µM duplex was used for capture of I-tailed RNA. Libraries were sequenced on the MinION using ONT R9.4 flow cells and the standard MinKNOW protocol script RNA001 or RNA002 recommended by ONT with one exception: Bulk phase raw files were collected for the first 2 h of sequencing and then standard sequencing was restarted and commenced for ∼48 h.

Base-calling and mapping reads to the yeast genome

We used Guppy Base-calling Software from Oxford Nanopore Technologies, Limited. Version 3.0.3 + 7e7b7d0 with the configuration file “rna_r9.4.1_70bps_hac.cfg” (Wick et al. 2019) was used for base-calling direct RNA traces. NanoFilt version 2.5.0 (Coster et al. 2018) was used for classification of passed reads. Reads classified as “pass” had a phredscore threshold of ≥7 and “failed” if <7. The alignments in Figure 7 were base-called using Guppy version 4.4.2 + 9623c16 with the same configuration file as above, but this did not result in significant changes to mapping or read counts. Passing reads were mapped to the sacCer3 yeast genome using minimap2 version 2.16-r922 (Coster et al. 2018; Li 2018) with the parameters -ax splice -uf -k10 -G2000.

Estimating tail lengths using Nanopolish

We used Nanopolish v0.10.2 to estimate homopolymer tail lengths. First the fast5 and fastq files are indexed using Nanopolish index v0.11.1 with the default parameters. Then Nanopolish poly(A) v0.11.1 was used to estimate the length of the homopolymeric signal using the default parameters. We used the event detection module from nanopolish GitHub branch r10 for estimating signal variance for poly(I) and poly(A) homopolymers.

DATA DEPOSITION

Fast5 files acquired from nanopores used in this study have been uploaded to the European Nucleotide Archive (https://www.ebi.ac.uk/ena/browser/home) under accession number PRJEB40734.

SUPPLEMENTAL MATERIAL

Supplemental material is available for this article.

COMPETING INTEREST STATEMENT

M. Akeson, L.M., and M.J. received reimbursement for travel, accommodation, and conference fees to speak at events organized by Oxford Nanopore Technologies (ONT). M. Akeson holds shares in ONT, is a paid consultant to ONT, has received research funding from ONT, and is an inventor on 11 UC patents licensed to ONT (6,267,872, 6,465,193, 6,746,594, 6,936,433, 7,060,50, 8,500,982, 8,679,747, 9,481,908, 9,797,013, 10,059,988, and 10,081,835). A patent application has been filed by the University of California naming J.V., L.M., M. Akeson, and M. Ares Jr. as inventors concerning the use of Cid1 for addition of modified nucleotides. J.Q.C. has no competing interests.