DragonRNA: Generality of DNA-primed RNA-extension activities by DNA-directed RNA polymerases

Abstract

RNA polymerases (RNAPs) transcribe DNA into RNA. Several RNAPs, including from bacteriophages Sp6 and T7, Escherichia coli, and wheat germ, had been shown to add ribonucleotides to DNA 3′ ends. Mitochondria have their own RNAPs (mtRNAPs). Examining reaction products of RNAPs acting on DNA molecules with free 3′ ends, we found yeast and human mtRNAP preparations exhibit a robust activity of extending DNA 3′ ends with ribonucleotides. The resulting molecules are serial DNA→RNA chains with the input DNA on the 5′ end and extended RNA on the 3′ end. Such chains were produced from a wide variety of DNA oligonucleotide inputs with short complementarity in the sequence to the DNA 3′ end with the sequence of the RNA portion complementary to the input DNA. We provide a set of fluorescence-based assays for facile detection of such products and show that this activity is a general property of diverse RNAPs, including phage RNAPs and multi-subunit E. coli RNAP. These results support a model in which DNA serves as both primer and template, with extension beginning when the 3′ end of the DNA is elongated with a ribonucleotide. As this DNA→RNA class of molecule remains unnamed, we propose the name DragonRNA.

Graphical Abstract

Open in new tabDownload slide

Introduction

The central dogma of molecular biology posits the transcription of DNA into RNA as a fundamental and critical process for gene expression, genetic inheritance, and life [1, 2]. Transcription is performed by DNA-directed RNA polymerases (RNAPs), which copy DNA into RNA beginning at promoter sequences [2–4]. Several studies have found RNAP from T7 bacteriophage (T7 RNAP) can perform promoter-less DNA-directed transcription [5, 6] and RNA-templated RNA synthesis [7–9]. T7 RNAP can also use individual short RNA oligonucleotides as both primer and RNA template in a single-base extension reaction, in which the RNA input forms a transient loop for the 3′ end to base-pair with a site in the input sequence, priming a single-base extension templated by the base upstream from the pairing [6]. Mitochondria, the organelle that produces and stores energy in eukaryotic cells [10], have their own DNA [11], which is transcribed by a nuclear-encoded mitochondrial-specific DNA-directed RNA polymerase (mtRNAP) [12]. mtRNAPs are single-subunit transcription polymerases very similar in structure to T7 RNAP [13–15]. The mtRNAP from slime mold Physarum polycephalum has also been found to perform nontemplated single-ribonucleotide extensions on RNA [16]. The mitochondrial RNAPs from Saccharomyces cerevisiae (Rpo41) [17] and humans (hmtRNAP) [15] are investigated here.

In addition to critically providing RNA for mitochondrial functions, mtRNAPs are the potential targets of interventions in cancer [18] and antiviral drug toxicity [19]. hmtRNAP mutations inhibit mitochondrial transcription and cause neurological disease and developmental delays [20]. Understanding the activities of mtRNAPs, and characterizing their modulation and roles in key cellular processes are important for understanding the impact of mitochondrial transcription, mtRNAP enzyme roles in key cellular processes, and mtRNAP activity, which may be valuable for specific biotechnological applications.

A few studies have shown that RNAPs can covalently link a ribonucleotide to the 3′ end of a DNA molecule. T7 RNAP can add ribonucleotides to the 3′ end of either an RNA or DNA molecule, which has been posited as an RNA- or DNA-editing mechanism [21]. RNAPs from bacteriophages T7 and Sp6 have been shown by G. Krupp [22] to use DNA oligonucleotides as primers for RNA synthesis, making hybrid DNA→RNA molecules (also observed with SPβ RNAP [23]). Furthermore, Krupp found that this T7 and Sp6 RNAP activity was robust with several different templates and generated DNA-primed and DNA-templated RNA extension with a “fold-back” structure [22]. Multi-subunit RNAPs wheat germ Pol II, in in vitro transcription reactions of SV40 virus [24], and Escherichia coli RNAP [23, 25, 26], have also been shown to covalently attach a ribonucleotide to the 3′ end of a DNA, often a single-stranded nicked DNA, and extend that 3′ end with an RNA chain. Together, these studies show that some nominally promoter-dependent and DNA-directed RNA polymerases, specifically from T7 and Sp6 bacteriophages, E. coli, and wheat germ, can extend the 3′ end of a DNA molecule with ribonucleotides. We here extend the analysis of serial DNA→RNA hybrids. We show that single subunit mitochondrial RNAPs have robust DNA-primed and DNA-templated RNA extension activities and can be used as models for formation of such DNA-capped RNAs. We also found corroborating results in assays with both phage (T7, Sp6, T3, and Syn5) and multisubunit (E. coli) RNAPs. Sequencing these DNA→RNA molecules, we show that the RNA portion is templated by the DNA input sequence, and that this activity happens in various input configurations, in both cis- and trans-priming, and on double-stranded DNA input with two strands of different lengths and on double-stranded DNA input containing a gap or nick.

As this DNA→RNA class of molecule has been recognized for several decades, it seems appropriate to provide a name for it, to provide clarity and distinction between this DNA→RNA molecule from other configurations (DNA:RNA hybrids [27], interspersed incorporations of RNA nucleotides by DNA polymerases [28], or RNA→DNA combinations such as those formed in Okazaki fragment production during DNA replication [29–31]). Thus, we propose to name these DNA→RNA combination molecules “DragonRNA,” with analogy to Dragons as a type of serial chimeric monster, and with homage to (and distinction from) Spiegelman’s RNA Monsters [32]. The possibilities raised by DragonRNA are important in designing in vitro transcription reactions, and should be considered in studying the nature of diverse RNA synthesis mechanisms in cells.

Materials and methods

Terminology

As our study describes a noncanonical activity of RNAPs, we use some terms to describe this activity in noncanonical ways. Therefore, we present definitions for these terms here in order to improve understanding of our study and analysis.

Input: Nucleic acid material intentionally added to the in vitro enzyme reaction at the start

Primer: Any nucleic acid serving as a starting point for covalent nucleic acid extension

Template: Nucleic acid material used by the RNAP to mold the new strand of RNA, with template complementarity determining the sequence of RNA that is produced

Product: Any molecule formed through an in vitro enzyme reaction and observed after the reaction has taken place

Extension: Any covalent continuation on the 3′ end of a DNA input/primer

Conditions: Buffer and temperature (unless temperature is otherwise specified, in which case only buffer) conditions for the reaction, as described in the “Materials and methods” section

Priming: Hybridization between the 3′ end of a nucleic acid molecule, caused by base pairing between the 3′ end and elsewhere in the input, which can take place either within the same molecule (via intra-molecular/cis base pairing via a loop structure) or between two distinct molecules (via inter-molecular/trans base pairing)

Expression and purification of polymerases

Recombinant Rpo41ΔN51 with His₆ tag was expressed and purified with modifications as previously described [33]. E. coli Rosetta (DE3) cells (Millipore Sigma 70954) expressing Rpo41 were resuspended in lysis buffer [50 mM Tris (pH 7.8), 100 mM NaCl, 10% glycerol, 1 mM phenylmethylsulfonyl fluoride (PMSF), and EDTA-free Pierce^™ protease inhibitor tablet (Thermo Fisher A32965)] and sonicated. The lysate was centrifuged at 31 000 rcf for 1 h at 4°C. After centrifugation, the supernatant was incubated with Ni-NTA resin (Qiagen 30210) equilibrated with buffer N [20 mM Tris (pH 7.8), 100 mM NaCl, 10% glycerol, and 1 mM PMSF] for 1 h. Ni-NTA resin was washed with buffer N supplemented with 10 mM imidazole (pH 8.0) and 1 mM β-mercaptoethanol (BME) and eluted with buffer N containing 250 mM imidazole (pH 8.0) and 1 mM BME. Eluent was concentrated and loaded onto the Superdex 200 size exclusion column (Cytiva) in buffer S [20 mM Tris (pH. 7.8), 1 M NaCl, 10% glycerol, 1 mM EDTA (pH 8.0), and 10 mM BME]. The peak fractions containing Rpo41 were pooled, concentrated, and aliquoted for storage at −80°C. Protein concentrations were determined based on the absorbance at 280 nm using a NanoDrop spectrophotometer (Thermo Fisher Scientific).

Recombinant hmtRNAPΔN150 with His₆ tag were cloned into pProEX HTb (Gibco BRL) and transformed into Rosetta (DE3) cells (Millipore Sigma). Cells were grown in LB media to 0.5 OD₆₀₀, induced with 0.2 mM of isopropyl-β-d-thiogalactoside (IPTG) at 16°C for 18 h, and harvested at 6000 rcf for 10 min at 4°C. Cells were resuspended in lysis buffer [25 mM Tris (pH 7.8), 300 mM NaCl, 10 mM BME, and EDTA-free Pierce^™ protease inhibitor tablet (Thermo Fisher A32965)] and sonicated. High salt buffer [20 mM Tris (pH 7.8) and 5 M NaCl] was added to bring up the final NaCl concentration in the supernatant to 750 mM. After stirring for 5 min., the lysate was centrifuged at 31 000 rcf for 45 min at 4°C. After centrifugation, the supernatant was subjected to PEI precipitation at [0.05%]_final and stirred for 5 min and centrifuged again at 31 000 rcf for 45 min at 4°C. The supernatant was loaded onto Ni-NTA resin (Qiagen), equilibrated with buffer N [20 mM Tris (pH 7.8), 500 mM NaCl, 10% glycerol, and 1 mM BME]. The resin was then washed with the buffer N containing 750 mM NaCl and eluted with buffer N supplemented with 1 M NaCl and 250 mM imidazole (pH 8.0). Eluent was dialyzed against dialysis buffer [20 mM Tris (pH 7.8), 100 mM NaCl, 10% glycerol, 0.5 mM EDTA (pH 8.0), and 1 mM dithiothreitol (DTT)] for 4 h and centrifuged at 3000 rcf for 15 min at 4°C to remove any precipitates. The dialyzed sample was loaded onto a Heparin column and ran on a gradient from 100 mM to 1 M NaCl in buffer H [20 mM Tris (pH 7.8), 10% glycerol, 0.5 mM EDTA (pH 8.0), and 1 mM DTT]. Fractions containing hmtRNAP were concentrated and loaded onto Superdex 200 size exclusion column in buffer S [20 mM Tris (pH 7.8), 500 mM NaCl, 10% glycerol, 0.5 mM EDTA (pH 8.0), and 1 mM DTT]. Peak fractions containing hmtRNAP were pooled, concentrated, and aliquoted for storage at −80°C. Protein concentrations were determined based on the absorbance at 280 nm using a NanoDrop spectrophotometer (Thermo Fisher Scientific).

Syn5 and KP34 RNA polymerase sequences were based on the sequences provided in [34] and [35], respectively. RNAPs were synthesized with an N-terminal His₆ tag. RNAP purifications were performed by Ni-NTA chromatography in 50 mM HEPES KOH pH 7.5, 100 mM NaCl, 10 mM DTT, 0.1% Triton X-100, and stored in 50% glycerol.

Oligonucleotides

Oligonucleotides were synthesized by IDT. DNA oligonucleotides were resuspended in TE, pH 8.0; RNA oligonucleotides were resuspended in TE, pH 7.4. All oligonucleotide names, descriptions, and sequences are shown in Table 1. This table shows all of the oligonucleotides used in this study and their IDs, sequences, and descriptions of their function.

Initial oligonucleotide designs were based on templates and primers used in a previous study of transcription by human mitochondrial RNA polymerase [36]. Extensions on template designs with nucleotides randomly chosen (using Python random.choice) to contain similar NTP proportions to the original oligonucleotides yielded AF-EG-2 DNA (“short,” 27 nt), AF-EG-3 DNA (“medium,” 37 nt), and AF-EG-7 (“long,” 70 nt), with sequences shown in Table 1. Some reactions, where noted, also contained a 20 nt RNA primer oligonucleotide, AF-EG-1.

Following initial results, we designed AF-EG-15, a 5′ fluorescein (FAM)-labeled DNA oligonucleotide of AF-EG-2, and made variants with 5′ FAM labels and sequences shown in Table 1. These variants all were 5′ FAM-labeled and included AF-EG-21, in which the Cs and Gs in AF-EG-15 were switched to Ts and As, respectively, and vice versa, to maintain the potential for base-pairing within the sequence, while changing the sequence and base-pairing strength. AF-EG-22 was a DNA oligonucleotide generated using Python random.choice with roughly the same base-proportions as AF-EG-15. AF-EG-30, was an oligonucleotide designed with three potential internal sites at which the two bases on the 3′ end could base pair.

We designed oligonucleotides containing the promoter sequence for RNAPs from bacteriophages T7 [37, 38] and T3 [38], based on oligonucleotides adapted from [37].

For testing priming, we made two sets of two 5′ FAM-labeled DNA oligonucleotides (AF-EG-23 and 24 had no Gs; AF-EG-25 and 26 had no Ts); each pair consists of one oligonucleotide predicted to prime via base pairing between the two bases on the 3′ end and another location in the sequence and one oligonucleotide where no such priming is predicted. To test requirements or preferences for priming location, we produced a series of 5′ FAM-labeled DNA oligonucleotides, including an oligonucleotide that contained three potential priming sites (AF-EG-62), and variations of AF-EG-62 without each priming site and each combination of priming sites (AF-EG-61, 63, 64, and 76–79).

For experiments where we sought to characterize the pool of products by sequencing without DNase treatment, we designed oligonucleotides containing 5′-phosphate to facilitate and focus an initial 5′ end capture. We designed AF-EG-65 with the sequence of AF-EG-2, and oligonucleotides AF-EG-73 and 74 containing unspecified bases to test intra- versus inter-molecular priming. We also designed 5′-phosphate AF-EG-85 as a primer to test the impact of double-stranded templates on DragonRNA activity and used templates of a single molecule with a hairpin loop containing a 0-base nick (AF-EG-81) and 3-base gap (AF-EG-80), and linear versions of this system with a third oligonucleotide providing double-strandedness (AF-EG-84) and templates containing a 0-base nick (AF-EG-83) and a 3-base gap (AF-EG-82). For oligonucleotides in a double-stranded DNA system for fluorescence gel assays, we designed two 5′ FAM-labeled DNA primers: AF-EG-66, a 5 nt DNA primer, and AF-EG-67, a 20-nt DNA primer. We designed templates for these 5′ FAM-labeled DNA primers, AF-EG-68, a template that could not self-prime, and designed AF-EG-69 to AF-EG-72, template oligonucleotides with hairpin loops and double-stranded regions with gaps of 20, 3, 1, and a 0 nt nick between the 3′ end of the primer and the start of the double-stranded region.

Oligonucleotide hybridization

Reactions containing an RNA oligonucleotide (AF-EG-1) with a potential DNA template were run following pre-annealing using 8 μM of the RNA oligonucleotide and 80 μM of the DNA oligonucleotide in 50 mM NaCl, heated to 95°C for 10 min, ramped down 0.09°C per second to 20°C to anneal, followed by a 4°C hold. We also performed this heat-cool process on the DNA-oligonucleotide-only and RNA-oligonucleotide-only conditions for the parallel reactions in those experiments. For these reactions, the annealed DNA/RNA oligonucleotides were added to the RNAP reaction at 200 nM.

For reactions with double-stranded, promoter-containing DNA templates (AF-EG-16 to AF-EG-20) and for the fluorescence assay with a double-stranded DNA system (AF-EG-66 to AF-EG-72), oligonucleotides were annealed following the same protocol, except that equal amounts of each strand were used, and the annealed product was added to the RNAP reaction at 1.25 μM.

For reactions with a double-stranded DNA system used for sequencing, oligonucleotides were annealed following the same protocol, except that these oligonucleotide complexes were annealed using 25 μM of the DNA primer AF-EG-85 and 50 μM of the template (which may be AF-EG-80, 81, 83, or 83), and in the versions with three oligonucleotides, 25 μM of AF-EG-84 was included in the same annealing reaction. The annealed product was added to the RNAP reaction at 2.5 μM.

In vitro mtRNAP assays with DNA oligonucleotides

We performed in vitro assays with hmtRNAP from human and Rpo41 from yeast using three conditions to confirm activity. One condition was similar to those from [36], denoted as “Lu conditions.” Lu reaction buffer was made up of 5 mM Tris–HCl, pH 7.5, 10 mM DTT, 20 mM MgCl₂, 0.1% Triton X-100, 10% glycerol, 75 μg/ml kanamycin, and 4 mM each rNTPs. These reactions were run at 22°C [36]. A second condition, denoted “KS,” consisted of reactions run under conditions typical to T7 RNAP in vitro reactions, following [7]. KS reaction buffer was made up of 40 mM Tris–HCl, pH 8.0, 80 mg/ml polyethylene glycol (PEG) 8000, 20 mM MgCl₂, 0.01% Triton X-100, 1 mM spermidine, 5 mM DTT, and 4 mM each rNTPs, and these reactions were run at 37°C [7]. The final condition, denoted “NEPol,” followed the NEB protocol for RNA synthesis (NEB M0251), which used a buffer containing 40 mM Tris–HCl, pH 7.9, 6 mM MgCl₂, 1 mM DTT, 2 mM spermidine, and 0.5 mM each rNTPs, and reactions were run at 37°C. Note that Rpo41 and hmtRNAP showed different activities, as evidenced by hmtRNAP generating DragonRNA on a slower time scale than Rpo41 under “Lu” and “KS” conditions, with activities appearing more similar under “NEPol” conditions.

Except in hybridized reactions as described above, 2.5 μM of the specified DNA oligonucleotide was added to each reaction.

Reactions were activated by addition of RNAP, with amount given per 10 μl reaction (3.26 μg Rpo41NΔ51, 3.75 μg hmtRNAPNΔ150, 2 μg T7 RNAP, 5 μg Sp6 RNAP, 0.4 μg KP34 RNAP, 2 μg Syn5 RNAP, 4 μg E. coli RNAP holoenzyme, and 3.5 μg E. coli RNAP core enzyme).

These reactions were run for the indicated time and quenched with equal volume of a quench buffer containing 100 mM EDTA, pH 8.0, and 0.4% sodium dodecyl sulfate (SDS). Following quench, the crude products were mixed with 75% (v/v) Gel Loading Buffer II (Ambion AM8547), the mixture heated to 95°C for 5 min, and run on a pre-run 10% or 15% TBE-Urea gel (Novex 10% EC6875BOX, 15% EC6885BOX). Gel percentages are noted with each figure, and we note that the 15% TBE-urea gels were generally preferred in the more recent experiments as we observed that the separation was clearer between the slower migrating bands and the input DNA oligonucleotides. Gels were stained with either SyBr Gold (Invitrogen S11494) or Helixyte Gold (AAT Bio 17595), which are chemically equivalent, to visualize nucleic acids by diluting 1:10 000 in TBE and incubating at room temperature for 15 min. Gels were imaged on Typhoon FLA 9500 (except figures with the black background, Fig. 1B and Supplementary Fig. S3A, imaged on the AlphaImager) and color-edited and contrast-enhanced using ImageJ. Note that FAM-labeled oligonucleotides appeared on the gel slightly higher up than expected based on ladder migration.

Purification of nucleic acid reaction products from RNAP reactions containing DNA oligonucleotides

RNA from the Rpo41 and hmtRNAP in vitro reactions used for nuclease digestions and for sequencing (Figs. 2, 4, 7, and 11), were extracted using saturated phenol:chloroform (1:1 v/v) followed by ethanol precipitation. Quenched reaction samples were diluted in a solution of 1 M ammonium acetate, 10 mM EDTA pH 8.0, 0.2% SDS, and 2 |$\mu$|l per sample of Glycoblue (Thermo Fisher AM9516). This was mixed with equal volume 1:1 phenol:chloroform, then centrifuged for 5 min at room temperature at 16 000 × g in pre-spun (30 s, 16 000 × g) Heavy Phase Lock tubes (QuantaBio 2302830). The supernatant was mixed with equal volume chloroform and then centrifuged in the Phase Lock tubes (5 min, room temperature, 16 000 × g). The aqueous phase was then transferred to clean siliconized tubes and mixed with three-times volume molecular-grade ethanol and incubated at −80°C for at least 16 h. The reactions were then spun at 4°C at maximum speed for at least 1 h followed by aspiration of the alcohol solution, washed with 80% ethanol, and dried prior to resuspension in TE pH 7.4.

Nuclease digestions of RNAP in vitro reaction products

In nuclease sensitivity assays, purified reaction products from the DNA-oligonucleotide-containing RNAP reactions were nuclease-digested with Turbo DNase (Invitrogen AM2238), RNase H (Epicentre H39500), and RNase A (NEB T3018-2). The Turbo DNase digestion was carried out using the provided buffer. RNase H digestion was carried out using a buffer containing 500 mM Tris–HCl, 1 M NaCl, and 200 mM MgCl₂. To avoid potential contamination with RNase A, reactions with Turbo DNase and RNase H were set up in parallel and put in the polymerase chain reaction (PCR) machine, and all reagents returned to the freezer prior to the tube of RNase A being taken out of the freezer. RNase A digestions were run with rCutSmart buffer (NEB B6004S). Digestions were run in parallel for 1 h at 37°C, followed by phenol:chloroform extraction and ethanol precipitation as provided above. Turbo DNase and RNase H purifications were performed in parallel, while RNase A reactions and purifications were performed after and independently to avoid RNase A contamination to the other samples. Digested-and-purified samples were mixed with 75% (v/v) Gel Loading Buffer II (Ambion), heated to 95°C for 5 min, and run on a pre-run 10% TBE-urea gel (Novex). Gels were imaged on Typhoon FLA 9500 and color-edited and contrast-enhanced using ImageJ. All nuclease digestions and corresponding gels were run in parallel with controls containing DNA and RNA oligonucleotides to confirm that each nuclease specifically digested the expected nucleic acid (data not shown).

Sequencing of hmtRNAP and Rpo41 DragonRNA

For sequencing analysis, the in vitro mtRNAP reaction products were purified using saturated phenol:chloroform extraction followed by ethanol precipitation as described above. The products from hmtRNAP reaction with AF-EG-7 and Rpo41 reaction with AF-EG-2 were digested with DNase I (NEB M0303) in effort to remove remaining input DNA oligonucleotides, and purified using phenol:chloroform extraction and ethanol precipitation prior to library preparation (Fig. 4). As the DNase step has the downside of removing input DNA oligonucleotides that were extended with RNA, we confirm their presence using a parallel strategy for library construction where DNase treatment was omitted to allow a somewhat smaller number of molecules to be sequenced completely from the 5′ end (all sequencing files are deposited at SRA BioProject ID: PRJNA1117892). For sequencing experiments in Figs. 7 and 11, we did not perform DNase digestion on the DragonRNA products prior to sequencing.

For RNA-sequencing of DragonRNA, we used a derivative of the TruSeq Small RNA sequencing protocol from Illumina (RS-200-0012), which utilizes adapter ligation (we call this protocol DruSeq for derivitized TruSeq). DruSeq utilized adapters and primers designed and adenylated in-house, and the addition of 20% PEG in the ligation steps, as has been shown to improve adapter ligation and reduce bias [39].

Oligo preparations

The 3′ adapter, AF-DG-35 was adenylated using the 5′ DNA adenylation kit (NEB E2610), according to manufacturer’s protocol. The adenylation reaction was purified with the Oligo Clean & Concentrator-5 (Zymo Research D4060) following manufacturer’s protocol, with two additional steps: (i) the spin column was spun empty prior to elution to remove wash buffer and (ii) the wash step was repeated for a total of two washes.

3′ adapter ligation

Sample RNA was mixed with 10 ng/μl final concentration adenylated adapter AF-DG-35 to a total volume of 2 μl and heated to 70°C for 2 min and cooled on ice. Then, 3 μl of total volume made up of 0.5 μl of T4 RNA Ligase 2, truncated K227Q (NEB M0373), 1× final concentration T4 RNA Ligase Buffer (NEB M0373), and 20% final concentration PEG-8000 was added to the RNA-adapter mix. The reaction was incubated at 25°C for 1 h, after which 0.5 μl of STP oligonucleotide (AF-DG-37) was added to a final concentration of 1.8 μM, followed by incubation at 25°C for 15 min.

5′ adapter ligation

The 5′ adapter (AF-DG-36) was heated at 70°C for 2 min and cooled on ice. Then, 0.5 μl of 20 μM AF-DG-36 (1.4 μM final concentration), 0.5 μl of T4 RNA Ligase 1 (NEB M0204), 1× final concentration T4 RNA Ligase Buffer (NEB M0204), 1 mM final concentration ATP, and 16.3% final concentration PEG-8000 were added to the reaction mix and incubated for 1 h at 25°C. The total volume of the reaction was 7 μl.

Reverse transcription (RT)

Six microliters of ligation reaction mix was transferred to a new tube and 1 μl of RT Primer (AF-DG-3 at 4 μM stock concentration) was added, and the mixture was heated to 70°C for 2 min and cooled on ice. 5.5 μl of RT reaction buffer containing 1 μl of SuperScript II (Thermo Fisher 18064014), 1× final concentration First Strand Buffer (Thermo Fisher 18064014), 500 μM final concentration dNTPs, and 8 mM final concentration DTT was added to the ligated RNA-RT primer mix, and the reaction was incubated at 50°C for 1 h.

PCR amplification

PCR amplification of the ligated, reverse-transcribed libraries was performed using NEBNext High Fidelity PCR kit (NEB M0541L). Briefly, 1× NEBNext High Fidelity PCR Master Mix was mixed with 300 nM Universal Primer (AF-DG-4), and 300 nM indexing primer from TruSeq Small RNA Library Preparation Kit (Illumina RS-200-0012). PCR was initiated at 98°C for 30 s for initial denaturation, followed by 16 cycles of 98°C for 10 s, 60°C for 30 s, and 72°C for 15 s, followed by final extension at 72°C for 15 min and 4°C hold.

Gel purification

Amplified libraries were separated on a 2% agarose gel and DNA in the range ∼122–200 nt was excised with a razor blade. Gel pieces were purified using Qiagen QiaQuick Gel Extraction kit (Qiagen 28 704), according to manufacturer’s protocol.

Sequencing

Purified libraries were pooled, denatured with sodium hydroxide, and 12 pM was loaded on Illumina MiSeq v3 with 150 cycles, single-end reads.

Bioinformatics

Sequencing data of DragonRNA reactions were analyzed using a k-mer counting algorithm, DragonMostCommon, to generate the highest-counts reads filtered to contain the input DNA sequence on the 5′ end and complementary sequences on the 3′ end. Note that there were numerous other structures present in the data with different starts, lengths, and configurations of homology to the input not shown in the text. Potential parsimony priming and base-pairing structure of reads were modeled using AgniAlign (FireLabSoftware Github). Sequencing data from DragonRNA experiments where the initial template had unspecified (“mixed”) base positions to determine inter- versus intra-molecular priming were analyzed using a matching algorithm NNcomplementarity and was visualized (with coding suggestions from ChatGPT [40]) using matplotlib and seaborn. Sequencing data from DragonRNA reactions with double-stranded input were analyzed using an algorithm to determine lengths of extension dsDragonExtension and were visualized (with help from ChatGPT [40]) using matplotlib and seaborn. Code has been deposited on the DragonRNA GitHub.

Results

Human and yeast mitochondrial RNA polymerases can produce extended molecules from a DNA template in reactions that do not require an RNA primer

In vitro reactions were performed with yeast mtRNAP (Rpo41) and human mtRNAP (hmtRNAP) with a DNA oligonucleotide template and RNA oligonucleotide primer pair based on previously published designs for hmtRNAP assays [36]. The RNA primer included an overhang so that RNA extension products would be 8 nt longer than the DNA input, in order to distinguish the RNA extension from the DNA input on the gel assay. These reactions showed new synthesis, as evident from the appearance of material that migrated more slowly than the starting material. Somewhat to our surprise, the appearance of new material was not dependent on the RNA primer. Further, we observed that the DNA input oligonucleotide band was depleted following the reaction. These phenomena, a higher molecular weight band in reactions with a DNA oligonucleotide and no RNA primer, as well as weakened DNA oligonucleotide input band after the reaction, were replicated across experiments under the same conditions with both Rpo41 and hmtRNAP and with three similar DNA oligonucleotides of different lengths: short 27 nt AF-EG-2, medium 37 nt AF-EG-3, and long 70 nt AF-EG-7 (Fig. 1). We hypothesized that this product was a serial DNA→RNA chimera molecule.

DNA primes RNA extension activity by hmtRNAP and Rpo41

We tested the hypothesis that the product was a DNA-primed, RNA-extended combination molecule using nuclease digestions of the products of the RNAP reactions. The analysis was carried out in mtRNAP reactions on standard DNA oligonucleotides (Fig. 2A), or with 5′ labeled DNA oligonucleotides (Fig. 2B and Supplementary Fig. S1). DNase digestion showed a near-disappearance of all bands on the gel, as expected for relatively short RNA extensions; we hypothesize that the faint bands that remain from Rpo41 reactions may be RNA, possibly from the 3′ end of the molecule. RNase H and RNase A digestions shift the products to lower sizes, with migrations comparable to or slightly slower than the input molecules (compared to Fig. 1C). The RNase H digestions of the reaction products provide evidence that the DNA→RNA combination molecule forms a structure in which the synthesized RNA is hybridized to the input DNA oligonucleotide (Fig. 2C).

An additional test for synthesis of DNA-primed, RNA-extended combination molecules used 5′ labeled DNA oligonucleotides in the RNAP reaction (Fig. 2B). In such an assay, we expect that extension on the input oligonucleotide itself would yield a slower-migrating band on the gel. Oligonucleotides were designed with two different 5′ labels (FAM and ATTO633) with the same oligonucleotide sequence as AF-EG-2 and were run in RNAP reactions under the same conditions with hmtRNAP and Rpo41. All of these reactions showed the fluorescent label shifting higher on the gel following the mtRNAP reactions with two different 5′ labels, thus yielding products that were covalently extended beyond the DNA input. Since the reaction conditions only included rNTPs, we hypothesized that the input molecules had been extended with RNA. This hypothesis was further supported by nuclease digestion, with the label bands becoming undetectable following treatment with DNaseI digestion and with bands returning to a size near the input DNA oligonucleotide when treated with RNase H and RNase A (Supplementary Fig. S1). The effect of the RNase H treatment in this assay corroborated that the sequences were hybridized to the DNA molecule (Supplementary Fig. S1).

Based on the inference that the slower-migrating bands represent chimeric 5′ DNA oligonucleotides extended on the 3′ end with RNA (and dragged on the gel), we use the term “DragonRNA” to describe these products.

Time courses of DragonRNA synthesis by mtRNAPs

To further characterize the activities of mtRNAPs in producing DragonRNA, we performed time courses of hmtRNAP and Rpo41. This hmtRNAP reaction forms one predominant DragonRNA extension band on the input DNA oligonucleotide (Fig. 3B). We found that Rpo41 generates DragonRNA within minutes in these reactions (Supplementary Fig. S2A), with longer extension products over the course of time (Fig. 3C). The slowest-migrating bands observed in the Rpo41 reactions appear longer than any predicted DragonRNA band would be if the RNA extension was generated via single-round self-templating (Fig. 3C, the longest such band would be twice the length of the input minus 3–5 bases [27*2- (3to5) = 49–51 nt]; we observe bands consistent with single extension, which we hypothesized to use the input as a template and to the 5′ end of that template (close to 50 nt). Multiple fainter slower-migrating bands (>70 nt), suggesting a more complex reaction can occur (see the sequencing analysis provided in Results Section IV entitled "RNA sequencing of mtRNAP-generated DragonRNA demonstrates templating on DNA input sequences", Fig. 4B, especially the bottom two examples, Supplementary File S2, especially S2I to S2L, S2O, and S2P, and Supplementary File S3). We also observed bands stained with total nucleic acid stain that lack a FAM fluorescent label that we hypothesize are a parallel set of RNAs initiated de novo on the DNA template and therefore lacking the 5′ label (Supplementary Fig. S2C and D).

RNA sequencing of mtRNAP-generated DragonRNA demonstrates templating on DNA input sequences

We had proposed above that the RNA extended on the 3′ end was templated from the DNA input sequences. We then tested this directly using RNA sequencing to understand the RNA generated in these DragonRNA reactions. We sequenced samples of DNA-oligonucleotide-initiated reactions with and without the RNA primer. This entailed dual-end-capture, cDNA synthesis library preparation and sequencing on DNase-treated, purified products from hmtRNAP [hours-scale (3 h) reaction with AF-EG-7, the longest DNA input tested, with and without the RNA primer; corresponding gel in Fig. 1B], and from Rpo41 [minutes-scale (∼1 min) reaction with AF-EG-2, with and without the RNA primer; corresponding gel in Supplementary Fig. S2B].

We utilize the sequencing data to build a picture of the hybrid DragonRNA molecules subject to sequencing. In parsing the DragonRNA sequencing reads from DNaseI-digested sequencing data, we note that DNaseI digestions of oligonucleotides are notably partial, especially for single-stranded DNA [41] or DNA:RNA hybrids [42]. Thus, an expectation was that a fraction of the molecules would remain intact, while others would lack varying portions of the 5′ DNA sequence. We analyzed the sequencing results to determine the sense/antisense composition and to identify the highest-counts species for products that (i) retain the input DNA sequence on the 5′ end, and (ii) where the 5′ sequences may be absent but the read contains a direct match or complementarity to the input DNA sequence. The former group would presumably indicate circumstances where the initial oligonucleotide may have escaped the DNase I, while the latter would indicate molecules that had been trimmed by DNase degradation. We interpret these sequencing reads as examples of products represented amongst the up-shifted material on the gel, while noting that the quantitative representations may not be precise; in particular, it remains possible some DNA→RNA combination molecules would be captured or represented poorly with sequencing.

The sequencing showed that the “sense” DNA template sequence was followed by antisense sequences from that template with both Rpo41 acting on AF-EG-2 and hmtRNAP acting on AF-EG-7. These results indicated that the mtRNAPs were using the DNA oligonucleotide input as a template for the RNA that was generated (Table 2). Since the sequencing showed reads with and without the 5′ DNA input, the aggregate data provide an opportunity to examine both DNA and RNA portions of the resulting molecules.

We visualized sequences filtered by requirements of specific length and relatedness to DNA input using AgniAlign (FireLabSoftware Github; AgniAlign does not modify the input sequences but simply provides a visualization of complementarity for single- or multi-pass polymerase self-priming reaction products) (Fig. 4; Supplementary Files S2 and S3). The 3′ RNA extensions shown here had regions complementary to the input DNA sequence, consistent with the RNAPs using the DNA as a template when making DragonRNA (hmtRNAP in Fig. 4A, Rpo41 in 4B). When looking at longer sequences (examples shown in the bottom two sequences in Fig. 4B), though these were less common, there were sequences that extend the initial product through apparent priming, transient unhybridizing, and repriming, of the original sequence and through potential priming of the RNA product (the latter producing downstream sequences matching the input DNA strand, shown in the bottom sequence in Fig. 4B). These longer reads may correspond to the slower-migrating bands on the gel assay in Fig. 3 that correspond to sizes longer than would be observed in DragonRNA extension templated in a single round of DNA priming and extension. In these products, the start of the RNA portion is complementary to the input DNA, which is acting as a template, but the RNA terminates beyond the 5′ end of the template DNA, thus providing an explanation for the slower-migrating bands observed on the gel. A more complete analysis of the many species observed in this sequencing analysis is shown in Supplementary File S3, including several species that appear in the sequencing results that are longer than would be expected from one round of priming and extension. Interestingly, we also found that both mtRNAPs often added variable numbers (between 2 and 7 nt) of rAs or rUs to the 3′ end of the DNA input prior to extending the RNA using the input DNA as a template (Supplementary File S3). We hypothesize that this could be due to a stuttering mechanism, in which homopolymer tracks (in this case rAs or rUs) are copied iteratively by the RNAP (alternatively, single bases could be added arbitrarily and potentially edited if they are not accurately templated based on the DNA input).

DragonRNA synthesis depends on ability for 3′ end of the DNA oligonucleotide to base-pair with the sequence

Previous studies of RNAPs have shown that 1–2 nt of complementarity between the primer and template were sufficient to initiate extension in T7 RNAP, in which both the primer and template were a single input RNA [6], and in E. coli RNAP in which the primer was an RNA dinucleotide and a DNA template [43]. To test our hypothesis that DragonRNA synthesis depends on priming between the 3′ end of the DNA input oligonucleotide and a complementary site elsewhere in the DNA sequence, we generated input DNA oligonucleotides with different priming capabilities (Fig. 5A), ran mtRNAP reactions with these oligonucleotides, and analyzed them via gel electrophoresis. We generated two sets of DNA oligonucleotides, two with a 3′ end predicted to disallow base-pairing, and two with 2-nt additions that allow them to base pair (AF-EG-23 has no Gs and cannot self-prime; AF-EG-24 has no Gs and can self-prime; AF-EG-25 has no Ts and cannot self-prime; AF-EG-26 has no Ts and can self-prime). Removing complementarity between the 3′ end and internal sequences of the DNA input eliminated extended DragonRNA synthesis in these reactions (oligonucleotides AF-EG-23 and AF-EG-25, Fig. 5B and C), although the Rpo41 reaction with AF-EG-23 has a faint and slightly up-shifted FAM-labeled band that may be indicative of occasional base addition (Fig. 5B, shown with FAM and nucleic acid images separated in Supplementary File S1). These oligonucleotides can be compared to counterparts (AF-EG-24 and AF-EG-26) where the 3′ ends are expected to allow internal priming at two loci each. Both AF-EG-24 and AF-EG-26 show DragonRNA extension activity with Rpo41 (Fig. 5B) and hmtRNAP (Fig. 5C). The nucleic acid-stained bands in Fig. 5B that do not appear in the FAM-label-imaging could be de novo RNA products and thus without DNA on the 5′ end. These data support a model in which DragonRNA synthesis depends on the ability of the DNA input oligonucleotide to base pair and prime between the 3′ end and another locus in the sequence, via either looping or priming with another molecule in the DNA input.

To assess whether there were unusual sequence characteristics of the initially-tested DNA oligonucleotides that allowed DragonRNA synthesis, we tested two additional DNA input oligonucleotides that share overall features with AF-EG-15 but have quite different sequences. One such oligonucleotide was AF-EG-21, where AF-EG-15 has been modified by switching Gs with As and Ts with Cs (Fig. 5A). The other oligonucleotide, AF-EG-22, maintained similar base-proportions as AF-EG-15. Each of the additional oligonucleotides allow some complementarity to the sequence with the two bases at the 3′ end; AF-EG-21 has three such complementary sites to the internal sequence while AF-EG-22 has two. Both AF-EG-21 and AF-EG-22 generated DragonRNA in in vitro reactions of both Rpo41 and hmtRNAP (Fig. 5B and C, respectively). The reaction with AF-EG-21 yielded a strong DragonRNA band that appeared on the gel assay to go to completion, as the input material band was not observed after the reaction with AF-EG-21. We hypothesize that this is due to a more stable terminal loop in AF-EG-21 than AF-EG-15, or due to stronger priming between the 3′ end and the internal site due to the guanine/cytosine base-pairing. As with AF-EG-15, the slowest-migrating band from AF-EG-22 was longer than any that would be made with a single round using the initial DNA oligonucleotide as primer and template alone (Fig. 5B and C). These results indicate that the synthesis of DragonRNA by mtRNAPs is a common feature with different single-stranded DNA oligonucleotide inputs.

Proximity and terminal homology impact priming site choice

To test a set of choices between different priming sites, we designed a family of input DNA oligonucleotides with multiple potential priming sites (Fig. 6A). The oligonucleotides had a maximum of three possible priming sites (AF-EG-62), with a series of variants mutating a combination of priming sites so they could no longer base-pair with the 3′ end of the oligonucleotide. We ran reactions with each oligonucleotide input using Rpo41 and hmtRNAP. We refer to the 5′-most site Site 1, the middle site Site 2, and the 3′-most site Site 3. Dropping priming Sites 1 and 2 individually and combinatorially did not change or remove the dominant DragonRNA product (Fig. 6B and C). In reactions with both Rpo41 (Fig. 6B) and hmtRNAP (Fig. 6C), any variants without Priming Site 3 decreased activity to the levels observed in reactions with the oligonucleotide variant AF-EG-76, which contained no priming site options between the 3′ end and elsewhere in the sequence. Thus, with these input DNA sequences, both RNAPs preferred the 3′-most complementary site to prime DragonRNA extension. We note that hmtRNAP did show a faint slower-migrating band in the reaction with AF-EG-76 (the input with no priming sites), indicating that a lack of priming did not completely abolish DragonRNA extension activity by hmtRNAP on this input. Therefore, even in some cases where the DNA input is not hypothesized to prime, the RNAP still covalently extends DNA with RNA, suggesting a possibility of residual mispriming or nontemplated extension, and indicating that priming ability of the input DNA is sufficient but not necessary to allow for hmtRNAP DragonRNA activity. This observation differed between the two RNAPs, as Rpo41 did not show this slower-migrating band (Fig. 6B).

Assessing cis- and trans-priming events in DragonRNA synthesis

Our results showed that priming between the 3′ end of the input DNA and elsewhere in the sequence was important for the observed DragonRNA synthesis reactions, and we wanted to know whether that priming was intra-molecular (a single DNA molecule looping and priming on itself) or inter-molecular (between two molecules). An oligonucleotide pool design with unspecified bases (a “barcode”) near the beginning facilitated this distinction, as we could determine whether the products resulting from DragonRNA synthesis on such templates had a strong tendency for extension bases to match their own barcode. Two such oligonucleotide pools were designed and tested in parallel, each with two unspecified bases toward the beginning of the sequence. With each oligonucleotide pool, we ran reactions with Rpo41 and hmtRNAP, and performed RNA-sequencing on the product. We then analyzed the sequencing to determine which unspecified bases were present in the primer region of the DragonRNA sequencing read, and whether the extended region contained bases complementary to those, and thus templated by the primer, or noncomplementary, and thus templated by another oligonucleotide in the reaction. We found that both hmtRNAP and Rpo41 with both templates tested showed DragonRNA extension regions antisense to the region of unspecified bases from the primer oligonucleotide and other oligonucleotides in the reaction, but that templating by the primer oligonucleotide occurred more frequently (consistent with intra-molecular or cis-priming) than priming by alternative oligonucleotides’ sequences (consistent with inter-molecular or trans-priming) (Fig. 7). This provides strong evidence that priming in cis is the prevalent mode of DragonRNA synthesis in these reactions. A signal consistent with trans-priming was seen at a lower level, with the ability of the polymerases to use a trans-priming configuration confirmed below (Results Section XII entitled "DragonRNA activity takes place ondouble-stranded templates").

Rpo41 acts rapidly to extend either DNA or RNA termini

DragonRNA synthesis reactions involve two types of nucleoside addition; the first base added to the initial input reflects ribonucleotide addition to a DNA 3′ end chemistry (2′-H, 3′OH), while subsequent additions represent addition of nucleosides to a canonical RNA 3′ end chemistry (2′OH, 3′OH). As the first is a noncanonical activity for a nominally DNA-dependent RNA polymerase, we considered the possibility that this first addition would be a rare (and potentially rate limiting) event. If this were the case, we would expect that providing a RNA-like 3′ terminus on an otherwise DNA-like input might greatly facilitate the reaction. To test this, we designed a chemically-synthesized DragonRNA oligonucleotide with the same sequence as AF-EG-15, in which the final base was a ribonucleotide adenine instead of a deoxyribonucleotide adenine. Visualization of products from the two templates was facilitated through two-color imaging (DNA terminus: green, RNA-terminus: red, pseudocolored as magenta in Fig. 8). Both templates were extended in three minute reactions with Rpo41, with no evident difference between the two. To observe potential differential usage, we ran reactions with progressively decreasing the Rpo41 concentration, finding comparable extension activities even when the reaction did not go to completion (Fig. 8), Thus we found no evidence that DNA termini form an inferior primer for the RNA polymerase.

Many RNAPs generate DragonRNA with equivalent templates

To test whether DragonRNA synthesis activity was specific to the tested mitochondrial RNAPs or common across a larger class of RNA polymerases, we ran DragonRNA experiments using a number of single-subunit RNAPs with varying degrees of relatedness to mtRNAPs. We generated a phylogenetic tree of the amino acid sequences of these single-subunit RNAPs in order to assess relatedness at the protein level (Fig. 9A, generated using [44–47]). We found that most of the RNAPs tested generated DragonRNA, including RNAPs from bacteriophages T7, T3, Sp6, and Syn5 (Fig. 9B; Supplementary Fig. S2C and D). We tested the oligonucleotides from Fig. 5A with T7 RNAP as well, and found the results to be similar to hmtRNAP and Rpo41 (Supplementary Fig. S4). Interestingly, RNAP from bacteriophage KP34 did not generate DragonRNA (Fig. 9B), even after 24h (Supplementary Fig. S5), despite the ability of this polymerase to synthesize RNA from a double-stranded template under these conditions (data not shown). It is possible that KP34 RNAP could generate DragonRNA under different conditions, or that DragonRNA synthesis is not an activity inherent to this enzyme, in which case, KP34 could be useful to produce RNA in a population without extension on input DNA, potentially generating a more pure RNA product in some applications.

Since we had found that many single-subunit RNAPs generate DragonRNA, we tested whether the multi-subunit, cellular E. coli RNAP can generate DragonRNA in vitro. We found that both E. coli holoenzyme and core RNAPs generated DragonRNA with AF-EG-15 (Fig. 9C), indicating that this activity is shared by at least one multi-subunit RNAP. We also found that E. coli core RNAP also generated DragonRNA with AF-EG-21 and AF-EG-22 and under “KS conditions,” which have been used for T7 RNAP in vitro reactions (see “Materials and methods” section) (Supplementary Fig. S6A and B).

To understand the technical aspects of the DragonRNA synthesis process and its potential impacts in physiology and biotechnology applications, it was of interest to determine how robust DragonRNA synthesis activity was across various of the RNAPs under diverse conditions. We tested different reaction conditions for T7 RNAP, Rpo41, and E. coli core RNAP to assess whether temperature and buffer conditions influenced DragonRNA synthesis activity (see “Materials and methods” section for condition details). E. coli core RNAP and Rpo41 showed DragonRNA synthesis under “KS conditions” (Supplementary Fig. S6B). Under “Lu conditions” for the buffer with a reaction temperature at 50°C (the highest hypothesized temperature in the mitochondria [48]), HiT7 (T7 RNAP that is stable at high temperatures) showed a dispersed electrophoretic pattern of DragonRNA, while Rpo41 showed a discrete band of the same size as at 22°C (Supplementary Fig. S6C). Time courses of T7 RNAP with AF-EG-15 under both “Lu” and “KS” conditions showed synthesis of DragonRNA, and under “KS conditions,” showed more RNA signal on the gel, which is not as strong under “Lu conditions” (Supplementary Fig. S6D). A set of conditions referred to as “NEPol” (see “Materials and methods” section) that have lower NTP concentration than the other conditions show DragonRNA activity (Supplementary Fig. S7, lanes 3, 6, and 8) and conditions with the addition of NaCl show DragonRNA activity (Supplementary Fig. S7, lanes 4 and 8) for Rpo41 reactions. Overall, these data show that a diversity of conditions allow DragonRNA synthesis activity.

Assessing T7 RNAP DragonRNA activity with a promoter-containing DNA template

Milligan et al. [37] described a valuable method for production of defined RNA products with T7 RNA polymerase that entails the use of a pair of oligonucleotides of different lengths to generate a partly double-stranded duplex with a promoter region that then transcribes into a single-stranded region from a template. Such templates also provide a potential starting point for DragonRNA and we assessed the consequences to the input DNA in these reactions, using both the single-stranded sense strand as input as well as the double-stranded promoter-containing template in parallel reactions (Fig. 10A). We ran these reactions with the sense-strand only (AF-EG-16 or AF-EG-18) as well as the annealed promoter-containing, double-stranded template mixtures (AF-EG-16 or 18 annealed to AF-EG-17 as described in the “Materials and methods” section). In addition to the T7-promoter-containing templates, we also ran these reactions with T7 RNAP with a non-T7-promoter-containing double-stranded template (with the T3 promoter instead of the T7 promoter). In these reactions, T7 RNAP produced a DragonRNA band in the sense-strand-only reactions under both “Lu” and “KS” conditions. In the reactions containing a double-stranded template, under “Lu conditions,” or under control conditions with a non-T7 promoter (T3 RNAP promoter), T7 RNAP made a very faint band of DragonRNA and showed a dispersed electrophoretic pattern at approximately the size of the promoter-driven RNA transcript product; under “KS conditions,” T7 RNAP did not make DragonRNA, but made strong bands of the expected, promoter-driven RNA transcript (Fig. 10B). These experiments raise the possibility that native template characteristics (double-stranded structure and presence of the promoter) are among the conditions that may limit DragonRNA synthesis in the complex environment of a functioning transcription complex. In at least this example, T7 RNAP reactions with a double-stranded, promoter-containing DNA template favored the promoter-driven RNA over DNA extension with RNA (Fig. 10B, lane 9), but both products can be observed under appropriate conditions (Fig. 10B, lane 4).

Molecular modeling of yeast Rpo41 generating a DragonRNA

To visualize the potential mechanism for DragonRNA synthesis, we modeled an Rpo41 complexed with input AF-EG-2 hairpin DNA oligonucleotide and an incoming rNTP. We based our model on the existing cryo-electron microscopy (cryo-EM) structure of an Rpo41 initiation complex with a promoter DNA and a 6-mer RNA primer [49]. Upon this model, we superimposed the 3′-terminal two A·T base pairs of the AF-EG-2 hairpin input DNA to the 3′-RNA/DNA heteroduplex of the cryo-EM structure. We note one particular Arginine residue of interest: in the yeast Rpo41 enzyme, this residue is Arg⁸²⁹; the residue is highly conserved among single-subunit RNAPs, including T7 RNAP (Arg⁴²⁵), and multi-subunit RNA polymerases, including E. coli RNA (Arg³⁵² of the β subunit), as well as with the DNA Pol I family [50]. The corresponding Arg in T7 RNAP (Arg⁴²⁵) is involved in catalysis and transition from initiation to elongation [50]. We model DragonRNA additions and the distance from the building nucleic acid and the Arginine of interest by Rpo41 (Supplementary Fig. S8A), T7 RNAP (Supplementary Fig. S8B), and E. coli RNAP β-subunit (Supplementary Fig. S8C). We hypothesize the 2′-OH of the terminal nucleoside in DragonRNA could interact with Arg⁸²⁹ and another amino acid, Asp¹¹⁶⁶ (located 7.2 Å from the 2′-terminus in the model; Supplementary Fig. S8A).

Further modeling (Supplementary Fig. S9A and B) also suggests that the hairpin DNA input oligonucleotide allows the conformation of the Rpo41 active site to mimic the Rpo41 conformation during the DNA template/RNA hybrid, and thus enables Rpo41 passing the very early stages of de novo RNA synthesis and entering the more stable RNA transcription reaction. Thus, this suggests that the conformation of Rpo41 at the start of DragonRNA synthesis could be more similar to that at the elongation stage of RNA synthesis, making the DragonRNA extension complex more stable and potentially more biochemically similar to RNA elongation than initiation.

Beyond the specific model of initial nucleotide addition, the structure is instructive in that the modeled DragonRNA occupies the same Rpo41-binding channel for the DNA/RNA heteroduplex in the promoter-containing initiation complex, offering an explanation for how promoter-driven RNA transcription can suppress DragonRNA synthesis.

DragonRNA activity takes place on double-stranded templates

As our initial DragonRNA reactions used single-stranded DNA oligonucleotides as the input material for DragonRNA reactions, we designed DNA oligonucleotide inputs with double-stranded character to test whether DragonRNA activity takes place on such templates. These designs entailed a set of oligonucleotide systems that consisted of (i) a DNA primer that could not self-prime and (ii) a pairing-capable template with substantial double stranded structure formed either with a hairpin loop or through annealing between the template oligonucleotide and a third oligonucleotide (Fig. 11A). In each case, we annealed the constituent oligonucleotides through heat-cooling prior to running the RNAP reactions. With either DNA oligonucleotide input design, Rpo41 generates DragonRNA (Supplementary Figs S10 and S11) templated based on the longer region of the template strand (Fig. 11). Such products can only be produced by a trans-priming event, and evidence the use of an essentially double-stranded template (with an effective nick or gap) as the substrate for the reaction.

In examining the products of these reactions, we observe that Rpo41 can extend DragonRNA past a region of gap between the primer and double-stranded region of the input material, indicating that Rpo41 is able to displace the nontemplate strand. Rpo41 extends DragonRNA to varying lengths when the alternative (template) strand’s double-stranded region is from a hairpin loop (Fig. 11D and E), potentially reflecting an ability of the hairpin loop to re-anneal while the Rpo41 is extending the DragonRNA, forming a structure that the Rpo41 must again displace (Fig. 11B and H). Rpo41 extends DragonRNA to the 5′ end of the input oligonucleotide when the template strand is linear and its double-stranded region is from a third oligonucleotide in the reaction (Fig. 11F and G). This indicates that the Rpo41 displaces the third oligonucleotide from the template strand and, being an independent molecule, it may have less propensity to re-anneal as the Rpo41 is extending the DragonRNA (Fig. 11C and I).

Discussion

We characterized the activities of several RNA polymerases in catalyzing a DNA→RNA extension, beginning with a noncanonical reaction in which the 2′ deoxy, 3′ hydroxyl end of the DNA is elongated with a ribonucleotide (Fig. 12B), thus extending the 3′ end of the DNA with RNA. We determined that this activity is common but not ubiquitous across RNAPs, including single-subunit mtRNAPs. These results extend previous findings with single-subunit phage RNAPs [21, 22], and multi-subunit cellular E. coli RNAP [25, 26]. Our results demonstrate that these activities are shared by additional phage DNA-directed RNA polymerases, from T3 and Syn5, and by mitochondrial RNA polymerases from human and yeast. Analyzing the activities of mitochondrial RNAPs in more detail, we also find that a variety of DNA oligonucleotides can support such reactions, including double-stranded DNA input, that extension can occur very quickly in varied polymerase reaction conditions, and that the RNA portion of such products are templated in large part by the input DNA sequence.

We have termed the resulting DNA→RNA combination molecule “DragonRNA” to provide clarity and distinction from other RNA/DNA chimeras and hybrids. The priming and initial 3′ rNTP addition are perhaps the most noncanonical aspects of this reaction (Fig. 12). We found that the DNA oligonucleotide 3′ end pairs with another site in the input DNA sequence followed by templated extension with ribonucleotides. After addition of the first ribonucleotide to the 3′ end of the DNA chain, the DragonRNA extension reaction shares characteristics with the classic RNA polymerization reaction [2–4, 51, 52]. Our results indicate that, for each input DNA oligonucleotide, the RNAPs often have one preferred priming site, as indicated by the consistent extension sizes observed in many of the reactions.

Sequencing allows a precise analysis of where DragonRNA synthesis reactions with the mitochondrial RNA polymerases finish on a DNA template. We find that many DragonRNA molecules terminate at the same point: at the 5′ end of the DNA template. We also found that longer extension reactions generated products that migrated more slowly, including longer nucleotide fragments than would have been possible from a single round of priming-and-templating with input DNA alone (gel bands Fig. 3B, sequencing Fig. 4B and Supplementary Files S2 and S3). Based on mobility and sequences present, we propose a multi-round repriming with the 3′ end, as diagrammed in Fig. 12. We found that such rounds of transient priming and repriming could be templated either on the same molecule or extended version of the original primer molecule (as exemplified in Fig. 12A), or can also take place via hybridization between two separate oligonucleotides in the reaction, followed by extension on one of them and templated by the other (both exemplified in Fig. 12A). We posit that DragonRNA sequentially serves as a primer and template in these reactions for successive rounds of RNA extension. Similar conclusions on phage polymerases from Sp6 and T7 were drawn by Krupp [22]. In both cases, the possibility of more complex reactions such as rolling circle (as proposed in [22]) or rolling hairpin replication resulting from such activities is intriguing.

Mammalian mtRNAPs may serve as primases in mitochondrial DNA lagging-strand replication, making short RNA primers that are extended with DNA on the 3′ end [53]. Additionally, ribonucleotides can be incorporated into genomic DNA by DNA polymerases during DNA replication [28]. DragonRNA is the inverse of the intermediate RNA→DNA hybrid molecules formed in the process of DNA replication, in which short RNAs prime Okazaki-fragmented DNA polymerization [29–31]; an intermediate molecule is thus made during DNA replication that builds DNA on the 3′ end of an RNA primer [54, 55]. The products are temporary RNA→DNA combination molecules with RNA on the 5′ end and DNA on the 3′ end [29–31, 54, 55]. Here, we show the inverse, in which the in vitro RNA-generated DragonRNA molecules are DNA 5′→RNA 3′ combination molecules.

On potential biological roles of DragonRNA

An extensive in vivo study to determine whether DragonRNA is synthesized in cells was outside the scope of this study; such analysis will be key to understanding the roles of the DragonRNA synthesis activities we describe across RNAPs. Combined with previous research [21, 22, 24–26], the work here shows the potential for formation of unusual DNA→RNA chimeras by enzymes present in diverse biological contexts that counter the textbook view of transcription. This work here shows the generality of this capability by numerous RNA polymerases, raising the question of whether such processes have not yet been observed because they are universally suppressed in vivo, or whether they are permitted or active under specific circumstances, or potentially broadly present and have simply been missed.

Mechanisms that might suppress the production of DragonRNA in the cell include ancillary factors that interact with a polymerase to increase specificity (e.g. E. coli DNA ligase has been shown to link 5′ DNA oligonucleotides with the 3′ end of RNA oligonucleotides [56]), a minimization of DNA 3′ ends in cells, complex structures that might engage or protect DNA 3′ ends, competition with canonical DNA polymerases, or yet-to-be-defined conditions that prevent the extension activities. Prevention of such activities could be critical for cells to avoid unwanted side products that would otherwise result in unwanted events at the chromosome level and potential production of inflammatory or damaging products. Free 3′ ends do occur in cells, with examples including the sites of nick- or overhang-producing DNA damage, processes coincident with Okazaki-fragmented DNA replication, viral or mobile element integration involving DNA breaks, or replication strategies that leave transient 3′ ends (e.g. [29, 31]).

Despite the existence of potentially suppressive mechanisms, there remains a real possibility that DNA extension with RNA does occur in the cell under specific circumstances–remaining to be identified. Certainly previous and extant methodologies would have often been insufficient to detect the DragonRNA observed in these experiments. Further, DragonRNA could be produced transiently yet still formed, play key roles, and/or represent molecular challenges to cells. RNAs that are primed by a nicked DNA (supported by Results Section XII "DragonRNA activity takes place on double-stranded templates" and [24]) would perhaps be a component of transcription induced upon DNA damage, allowing an additional functional modality for potential repair and/or expression mechanisms [57, 58]. These processes leave transient single-stranded DNA 3′ ends that may be susceptible to potential mitigation or interference by DragonRNA synthesis. If there are loci in the genome prone to events that yield available single-stranded 3′ ends that can be extended with RNA as a repair mechanism, the resulting DragonRNAs could serve either as a functional marker or productive effector in cell states associated with specific DNA damage. It is further conceivable that there is a class of RNAs that are generated via priming and extension, potentially iteratively. Future in cellulo or in vivo studies could follow up on work of DragonRNA to study these conditions and investigate the presence of DragonRNA in cells.

Dragons doing good? Potential for DragonRNA in biotechnology

We consider the potential options for DragonRNA synthesis to be leveraged for technological and therapeutic applications. RNA is extremely useful as a potential regulator of gene expression as well as in biochemical, synthetic biology, and medical applications [59–62]. Despite their remarkable utility, the use of RNAs for genetic, immunological, and potentially therapeutic intervention is often limited by natural mechanisms that degrade incoming RNA in cells and organisms [62, 63], with efficacy and consequence determined by mechanisms including innate immune sensing and response pathways [59]. In cells, a diversity of 5′ caps are added to RNAs to protect against degradation, most canonically 7-methylguanosine, but also including a diversity of other caps including adenine-containing cofactors [64, 65]. Certain synthetically introduced 5′ caps on RNAs have similarly shown promise for increased translation and improved stability [66]. DragonRNA is effectively 5′ DNA-capped RNA. While we don’t know the capabilities of DragonRNA in vivo, there is certainly potential for such capabilities to include downstream events that would provide value (either translation or modulation of gene expression) and for the DNA cap to provide either stability or application–favorable interaction with innate immune mechanisms.

In considering potential applications of DragonRNA, key questions are whether DragonRNA can be active in translation or in hybridization to potential target mRNAs. A translated DragonRNA (potentially utilizing an internal ribosome entry site [67]) could make a useful tool for RNA-based vaccines or therapeutics. If DragonRNA shows promise for such applications, RNAPs could potentially be engineered to specifically have DragonRNA activity. The mechanistic understanding of DragonRNA synthesis presented here provides an opportunity for further research on noncanonical activity by RNA polymerases and in the mitochondrial transcription systems.

Important for biotechnology applications, we showed that DragonRNA activity is robust in reactions where the input DNA oligonucleotides are double-stranded, with Rpo41 able to displace a double-stranded region to extend DNA with RNA (Fig. 11B and C). This result indicates a helicase activity of Rpo41 that would allow for DragonRNA activity on effectively double-stranded templates, including dsDNA templates (such as plasmids) in vitro, and cellular environments where DNA is nicked or gapped. Furthermore, this ability of the RNAP could also be a useful synthetic tool, allowing flexibility in the design of molecules synthesized in vitro (e.g. programming defined templates by addition of designed molecules to in vitro transcription reactions) or in vivo (e.g. utilizing targeted nickase activity). This could design a DragonRNA with user-defined DNA and RNA sequences, enabling a variety of applications.

Conclusion

The many noncanonical RNAP activities give interesting insight into the complex nature of biology, the central dogma, the interactions between DNA, RNA, and protein. DragonRNA furthermore lends itself to studies of the transition from the RNA-Protein World to the DNA World governed by the central dogma focused on noncanonical activities of RNA polymerases. Functional RNAs are known with a variety of 5′ structures. E. coli RNAP has been shown to extend various nucleotide and nucleoside derivatives with RNA [68], and hmtRNAP and Rpo41 have similarly been shown to extend NAD+ and NADH with RNA [64], resulting in molecules with a diversity of 5′ chemical structures. Perhaps DragonRNA provides an additional population of 5′ capped RNAs in certain biological contexts, fulfilling roles as either coding or noncoding nucleic acids. As discussed, perhaps a functional DragonRNA could additionally fulfill needs in RNA-based biotechnological intervention. Given previous research identifying DNA→RNA extension, and our findings that this is relatively common across many RNAPs, perhaps some transcriptions entail a priming reaction such as those shown here and previously. This poses questions about RNAPs’ activities, on whether all transcription is truly a de novo process, the RNA/DNA World Hypotheses, and about the evolution of genetic inheritance at the dawn of life.

Acknowledgements

We thank Dongxian Wu and Katherine Marks for help with protein purification. We thank Shoshana Williams, Tanner Jensen, and Nora Trapp for guidance on the work and help with figure design. We thank Usman Enam for help with phylogenetic tree generation. We thank Usman Enam, Ivan Zheludev, Karen Artiles, Dae-Eun Jeong, Dan Herschlag, Michael Bassik, Anne Brunet, Anne Villeneuve, Julie Baker, and Karla Kirkegaard for their advice, experimental and analytical guidance, and feedback on the work. We thank members of Stephen Montgomery’s and Andrew Fire’s laboratories for general guidance and feedback on this work.

We would like to acknowledge the Muwekma Ohlone community, the traditional stewards and caretakers of the land E.G., D.G., N.J., S.B.M., and A.Z.F were on while performing this work, the Karankawa, Ishak, Esto’k Gna, and Akokisa communities, the traditional stewards and caretakers of the land J.P. and Y.W.Y. were on while performing this work, and Massachusett, Wampanoag, Pawtucket, and Agawam communities, the traditional stewards and caretakers of the land B.R. was on while performing this work. Indigenous peoples are present and thriving despite the occupation of their ancestral lands and we affirm Indigenous sovereignty, history, and experiences, and accept our continued role in this story of colonization.

Author contributions: Emily Greenwald (Conceptualization [equal], Data curation [equal], Formal analysis [equal], Investigation [lead], Methodology [lead], Project administration [equal], Software [equal], Validation [lead], Visualization [lead], Writing—original draft [lead], Writing—review & editing [lead]), Drew Galls (Methodology [supporting], Writing—review & editing [supporting]), Joon Park (Investigation [supporting], Methodology [supporting], Resources [equal], Writing—original draft [supporting], Writing—review & editing [supporting]), Nimit Jain (Investigation [supporting], Methodology [supporting], Writing—review & editing [supporting]), Stephen B. Montgomery (Project administration [supporting], Resources [equal], Supervision [equal], Writing—review & editing [supporting]), Bijoyita Roy (Resources [equal], Writing—review & editing [supporting]), Whitney Yin (Formal analysis [supporting], Investigation [supporting], Resources [supporting], Supervision [equal], Visualization [supporting], Writing—original draft [supporting], Writing—review & editing [supporting]), and Andrew Fire (Conceptualization [equal], Data curation [equal], Formal analysis [equal], Project administration [equal], Resources [equal], Software [equal], Supervision [lead], Visualization [supporting], Writing—review & editing [equal])

Supplementary data

Supplementary data is available at NAR online.

Conflicts of interest

N.J. is currently an employee of Altos Labs Inc. S.B.M. is an advisor to MyOme and PhiTech Bio. S.B.M. consults for BioMarin, Character Bio and Valinor Therapeutics. B.R. is currently an employee of Moderna Inc.

Funding

Grants from National Institutes of Health (NIH) (R35GM130366 to A.Z.F.; R01MH125244 and R01AG066490 to S.B.M.; R01GM145925 to Y.W.Y.) are acknowledged for support of the work. E.G. was funded by the National Science Foundation Graduate Research Fellowship Program grant DGE-1656518, R01MH125244, and R01AG066490. D.G. was funded in part by NIH-T32HG000044-26. J.P. was funded by the Jean B. Kempner postdoctoral fellowship. Funding to pay the Open Access publication charges for this article was provided by the R35GM130366 to A.Z.F.

Data availability

Sequencing data from Fig. 4 that included a DNase digestion step and Figs. 7 and 11 that omitted the DNase digestion step are deposited on SRA BioProject ID: PRJNA1117892, and the table describing those data is included in Supplementary Table S1.

References

Author notes