Protein-side-chain-templated DNA synthesis in an antiphage reverse transcriptase: implications for noncoding RNA function

1 Introduction

Noncoding RNAs (ncRNAs) occupy a central and expanding role in prokaryotic physiology, extending from small regulatory RNAs that tune transcription and translation (Storz et al., 2011) to the CRISPR RNAs that direct sequence-specific adaptive immunity against mobile genetic elements (Barrangou et al., 2007). The most rapidly growing arena for prokaryotic ncRNA function is antiphage defense. Systematic surveys of microbial genomes have uncovered dozens of previously unrecognized defense systems clustered in defense islands (Doron et al., 2018; Gao et al., 2020), establishing that the immune arsenal of bacteria is far larger and more mechanistically heterogeneous than the restriction–modification and CRISPR–Cas paradigms alone implied (Bernheim and Sorek, 2020; Hampton et al., 2020; Georjon and Bernheim, 2023), spanning chemistries as varied as NAD⁺ depletion (Garb et al., 2022). Many of these systems operate through abortive infection, in which an infected cell arrests or dies before the phage completes its replication cycle, protecting the surrounding population (Lopatina et al., 2020).

Reverse transcriptase (RT)-based systems form a distinct and ancient component of this arsenal. Bacterial RTs are phylogenetically diverse and predate the eukaryotic retroelements that descend from them (Xiong and Eickbush, 1990; Toro and Nisa-Martínez, 2014), encompassing group II introns, retrons, diversity-generating retroelements, and numerous uncharacterized groups (Simon and Zimmerly, 2008; Kojima and Kanehisa, 2008; Zimmerly and Wu, 2015). Within this diversity, retrons (Millman et al., 2020; Bobonis et al., 2022) and the broader unknown-group/abortive-infection (UG/Abi) family (Mestre et al., 2022) have been recognized as antiphage modules in which an RT acts together with an ncRNA. The defense-associated reverse transcriptases (DRTs), first defined systematically among these elements (Gao et al., 2020), have since become a focus of mechanistic and structural work that has exposed an unexpected variety of DNA-writing chemistries: DRT2 assembles a toxic gene de novo from its ncRNA (Wilkinson et al., 2024), DRT9 synthesizes long poly(A)-rich cDNA (Song et al., 2025; Tang et al., 2025), and DRT4 produces template-independent single-stranded DNA of semirandom sequence (Wang et al., 2026). In each case the associated ncRNA has emerged as a functional participant in synthesis rather than a regulatory accessory.

Against this background, Deng et al. (2026) provide the first near-atomic-resolution structural description of a prokaryotic ncRNA functioning as a direct, sequence-specific polymerization template within an antiphage ribonucleoprotein (RNP) complex. The DRT3 system synthesizes alternating poly (GT/AC) double-stranded DNA (dsDNA) through a mechanistically asymmetric division of informational labour. One product strand is templated by the ncRNA through canonical Watson–Crick base pairing; the complementary strand is synthesized entirely through the geometry of a protein active site, with no nucleic acid template involved. Two distinct modes of sequence specification therefore operate within a single complex.

This informational asymmetry is the subject of the present Opinion. I argue that DRT3 is best understood not as one more antiphage RT but as a structurally resolved instance in which the sequence information specifying one product strand has been relocated from RNA into protein, and that this relocation reframes three questions for the RNA research community: the functional repertoire that prokaryotic immune ncRNAs can occupy, the way informational roles are partitioned between RNA and protein within a single machine, and the directionality and evolutionary stability of that partitioning. Consistent with the conventions of the Opinion format, the structural and biochemical findings of Deng et al. (2026) are first summarized without interpretation; the implications are then developed as the substantive contribution of this article.

2 The DRT3 ribonucleoprotein complex

Deng et al. (2026) determined two cryo-electron microscopy (cryo-EM) structures of the Escherichia coli GF-3 DRT3 complex (EcDRT3), a resting (−dNTP) and an elongating (+dNTP) state, each resolved to 2.6 Å. The complex assembles as a D3-symmetric 6:6:6 RNP containing six copies each of Drt3a (a class 2 unknown-group reverse transcriptase), Drt3b (a class 1 unknown-group reverse transcriptase), and the ncRNA. The asymmetric unit is a 1:1:1 Drt3a–Drt3b–ncRNA protomer; three protomers form a trimer, and two trimers dimerize tail-to-tail to yield the hexameric assembly. Intra-trimer stability is maintained by extensive Drt3a–Drt3b and Drt3b–Drt3b contacts, including a cyclical salt-bridge network between Glu389 and Arg390 of adjacent Drt3b subunits, while the inter-trimer interface buries approximately 1,768 Ų of surface area stabilized by at least seven conserved residues. The resting and elongating states are globally similar, with a root-mean-square deviation of approximately 0.302 Å between core Drt3b hexamers, localizing the principal conformational differences to the active sites. The poly (GT/AC) dsDNA product is resolved around the exterior of the complex. Co-purification of the ncRNA with the protein complex under native conditions, confirmed by TBE–urea PAGE and small-RNA sequencing, together with abolition of antiphage defense upon ncRNA deletion, establishes the ncRNA as a stoichiometric and functionally essential component. The ncRNA comprises approximately 130 nucleotides folded into four stem-loops (SL1–SL4); the single-stranded region between SL1 and SL4, specifically the A¹⁰⁸CACAC¹¹³ motif, serves as the direct template for poly (GT) strand synthesis by Drt3a.

3 The noncoding RNA as a polymerization template

Positioning of the ACACAC template motif within the Drt3a active site is mediated by conserved protein–RNA interactions that extend beyond non-sequence-specific backbone contacts. Tyr419 of Drt3a engages the G98–C107 base pair of SL4 through π–π stacking, anchoring the upstream boundary of the single-stranded template region, while Phe402 stabilizes the U19–A116 base pair of SL1 through an analogous aromatic interaction, anchoring the downstream boundary. Arg335 contacts C113, the terminal base of the ACACAC motif, through a cation–π interaction. Within the templating register, A108 acts as the primary templating base and is positioned by a hydrogen bond between its 2′-OH and the backbone carbonyl of Gly164, a residue of the canonical RT-4 motif (GxxxGxxxS) conserved across reverse transcriptases. A108 and C109 serve as the templating bases for alternating GT synthesis, while A110–C113 stabilize the nascent product strand. A five-nucleotide product (TGTGT) is resolved in the resting state and all six nucleotides in the elongating state, with the product strand extending to the final template base, consistent with a post-translocation state in which strand separation rather than nucleotide incorporation is rate-limiting. A β-hairpin extension unique to the Drt3a thumb domain (residues 323–338), positioned immediately downstream of C113, is proposed to mediate this dissociation step and to enable processive, cyclic dinucleotide addition. The downstream U114–A115 region is conformationally flexible and resolved only in the elongating state, where the C113–A116 inter-phosphate distance contracts from 15.4 Å to 10.3 Å, facilitating the register shift that positions C109 as the next templating base. Single-nucleotide mutagenesis confirms the functional indispensability of the ACACAC sequence: A108G (GCACAC) and C109T (ATACAC) substitutions abolish DNA synthesis in vitro and in E. coli and eliminate antiphage defense against phage T1.

4 Protein-templated synthesis of the complementary strand by Drt3b

Drt3b adopts a canonical right-hand reverse transcriptase fold but possesses structural features that preclude template binding. The template-binding channel is physically blocked by the Drt3b C-terminal segment (residues 645–650) and an extended Drt3b-specific internal loop (residues 112–169) that sandwiches the C-terminal segment against the path normally occupied by a template strand, while the conserved N-terminal Glu26 occupies the spatial position of the templating nucleobase. Comparative alignment with the Geobacillus stearothermophilus group II intron reverse transcriptase (PDB: 6AR1; Stamos et al., 2017) renders this exclusion explicit: the template-entry channel that is sterically open in canonical reverse transcriptases is fully occluded in Drt3b. Despite the absence of a nucleic acid template, Drt3b produces a defined poly (AC) sequence. The nascent cDNA strand adopts a backbone kink that compresses three bases (dC15–dA16–dC17) into the envelope normally occupied by two, enabling base-specific contacts between protein side chains and six consecutive positions of the nascent strand (dC13–dA18). Glu26 forms two hydrogen bonds with the N6 amine of the incoming dATP, discriminating it from dGTP and dTTP; Arg253 provides a cation–π interaction enforcing purine preference at the dA site; Gly248 sterically excludes dG; and the dC position is enforced by hydrogen bonds from Arg253 and Glu26 to the Watson–Crick edge of the preceding dC, with steric exclusion of purines completing the specificity. The E26A and E26Q variants confirm that Glu26-mediated hydrogen bonding is essential: E26A produces severely truncated cDNA, and both variants permit dG misincorporation at dA sites at approximately 19–22% upon dGTP addition, while dC fidelity is maintained. Synthesis is initiated by Tyr650, a C-terminal tyrosine whose hydroxyl acts as a priming nucleophile, a mechanism shared with AbiK, AbiP2, AbiA, and DRT9 and confirmed by the Y650F variant, which abolishes covalent adduct formation and eliminates antiphage defense in vivo.

5 Evolutionary conservation of the RNA–protein interface

Phylogenomic analysis of 1,232 DRT3 representatives distributed across at least 20 bacterial phyla shows that both the (AC)₃ ncRNA template motif and the key Drt3b templating residues Glu26 and Arg253 are co-conserved across this pan-bacterial distribution. Topological congruence between the individual Drt3a and Drt3b phylogenetic trees supports co-evolution of the two reverse transcriptase subunits and, by extension, with the ncRNA. Two DRT3 systems sharing only 32% amino acid identity, EcDRT3 from E. coli GF-3 and Ec2DRT3 from E. coli ECOR12, both assemble the 6:6:6 RNP and produce constitutive poly (GT/AC) dsDNA, indicating that functional RNA–protein interface integrity is maintained even at low sequence identity. The universal conservation of the poly (GT/AC) product across DRT3 homologs indicates that both the ncRNA template information and the protein-encoded sequence information have been maintained by purifying selection.

6 Discussion

The findings summarized above are most consequential, in my assessment, not for what they reveal about one defense system but for what they demonstrate about the partitioning of sequence information between RNA and protein. DRT3 resolves, at near-atomic resolution, a state in which the information specifying one strand of a nucleic-acid product resides in an ncRNA template while the information specifying the complementary strand resides in the geometry of a protein active site. This dual-mode specification within a single RNP is the conceptually decisive feature of the system, and it requires careful statement in relation to the central dogma. Crick’s formulation concerned the residue-by-residue transfer of sequence information and asserted only that such information cannot flow out of protein back into nucleic acid or protein (Crick, 1970); it did not prohibit a protein from specifying a nucleotide sequence through steric and electrostatic complementarity. DRT3 is therefore not a violation of the central dogma. It is, more precisely, a structurally documented case in which sequence specification that an ncRNA could in principle supply has instead been embedded in protein, and that distinction is what makes the system informative rather than merely surprising.

The DRT3 ncRNA is not a passive scaffold. The conformational change that contracts the C113–A116 inter-phosphate distance during elongation couples ncRNA geometry directly to the catalytic cycle, a mode of ncRNA-directed kinetic control over polymerization that is distinct from the static scaffolding attributed to most prokaryotic immune ncRNAs. The closest functional analogy is telomerase, the archetypal ribonucleoprotein in which an internal RNA template directs reiterative DNA-repeat synthesis (Greider and Blackburn, 1985; Nakamura et al., 1997) and in which a thumb-domain element governs product–template dissociation as the rate-determining step (He et al., 2021); the Drt3a-specific β-hairpin extension appears to perform an analogous role within an immune context. This places DRT3 within a continuum of RNA-templated DNA polymerases whose deepest roots lie in the bacterial group II intron RTs from which retroviral, retrotransposon, and telomerase RTs descend (Xiong and Eickbush, 1990; Stamos et al., 2017). What distinguishes the DRT3 ncRNA within that continuum is the precision of its template: single substitutions at A108 or C109 abolish both synthesis and defense, an intolerance at defined positions that exceeds what has been documented for most prokaryotic ncRNA–protein interfaces and places the DRT3 ncRNA among the most rigorously characterized functional ncRNA templates described to date.

The complementary strand is, for the RNA research field, the more consequential half of the system. That Drt3b specifies a defined poly (AC) sequence through side-chain chemistry alone, with template binding structurally excluded, means that the sequence information for this strand is held in protein rather than in nucleic acid. This is a qualitatively different arrangement from that of the other characterized ncRNA-dependent RT defense systems. In retrons, the RT copies its ncRNA into the hybrid multicopy single-stranded DNA that the system uses for sensing and signalling (Millman et al., 2020; Bobonis et al., 2022); in DRT2, the ncRNA is the template from which a de novo toxic gene is reverse-transcribed (Wilkinson et al., 2024); and in DRT9 the ncRNA templates a long homopolymeric cDNA (Song et al., 2025; Tang et al., 2025). In every one of these systems the RNA remains the carrier of product sequence. DRT3 is the only characterized case in which the determination of one strand’s sequence has been transferred to protein structure. Whether that transfer reflects internalization of an ancestral ncRNA template into the Drt3b active site over evolutionary time, or independent acquisition of specificity within the class 1 UG-RT clade, is unresolved; either trajectory makes DRT3 a uniquely tractable system for studying how informational roles migrate between the two macromolecular classes.

Viewed across the DRT family, these systems are converging on a coherent principle: antiphage RTs constitute a repertoire of noncanonical DNA-writing machines that exploit ncRNA-encoded information in mechanistically diverse ways. Template-free synthesis (DRT4; Wang et al., 2026), RNA-templated homopolymer synthesis (DRT9; Song et al., 2025; Tang et al., 2025), ncRNA-templated gene assembly (DRT2; Wilkinson et al., 2024), and now protein-templated dinucleotide-repeat synthesis (DRT3; Deng et al., 2026) span a spectrum from fully RNA-encoded to substantially protein-encoded product specification, with DRT3 marking the protein-internalized extreme. The physiological logic that unifies them is abortive infection: each produces an aberrant nucleic-acid species whose accumulation is incompatible with continued phage propagation, arresting or killing the infected cell for the benefit of the population (Lopatina et al., 2020; Millman et al., 2020).

The conservation data support an explicit evolutionary reading. Co-conservation of the (AC)₃ ncRNA motif together with the Glu26 and Arg253 templating residues across at least 20 phyla, and the topological congruence of the Drt3a and Drt3b phylogenies, argue that the asymmetric architecture of DRT3 is not a transient intermediate but a stable, functionally essential solution maintained by purifying selection over deep evolutionary time. This frames a tractable and, in my view, high-value question: do intermediate states persist within the broader UG-RT family in which an ncRNA template and partial protein-based contacts jointly specify a product strand? The phylogenetic breadth of bacterial RTs (Xiong and Eickbush, 1990; Toro and Nisa-Martínez, 2014; Zimmerly and Wu, 2015) and the defined boundaries of the UG/Abi family (Mestre et al., 2022) make a systematic reconstruction feasible, and pairing it with structural characterization of candidate intermediates would test directly whether protein internalization of template information arose gradually and whether it has occurred more than once.

Several mechanistic questions remain open and define near-term priorities. The downstream effector function of the poly (GT/AC) dsDNA product is unresolved. The product may act as a molecular decoy that titrates phage-encoded DNA-binding proteins, by analogy with the sequestration model proposed for the poly(A)-rich output of DRT9 (Song et al., 2025; Tang et al., 2025); it may adopt non-B-form conformations, such as slipped-strand structures, that serve as recognition signals for a host sensor; or it may be processed by an as-yet-unidentified nuclease or effector activated by the phage-encoded trigger ST61. Discriminating among these models will require identification of the effector and its partners, an objective approachable through structural prediction, co-immunoprecipitation coupled to proteomics, and genomic-neighbourhood analysis of DRT3 loci, the last now routine given systematic defense-system annotation tools (Payne et al., 2022).

The broader significance of Deng et al. (2026) extends beyond DRT3. Bacterial antiphage systems are increasingly recognized as a shared community resource, mobilized between strains by horizontal gene transfer (Bernheim and Sorek, 2020), and as the evolutionary source of conserved eukaryotic innate-immune modules, from nucleotide second-messenger signalling to the cGAS–STING pathway (Cohen et al., 2019; Wein and Sorek, 2022). A mechanism that encodes nucleotide-sequence information in protein and writes it into DNA without a template adds a genuinely new capability to this catalogue and to the synthetic-biology toolbox, where template-independent and protein-templated DNA synthesis could be repurposed for sequence-defined writing. For the RNA research community in particular, the value of DRT3 lies in providing a structurally defined reference point for a question that has been difficult to study directly: how, and how stably, sequence information is partitioned between RNA and protein. The deposited EcDRT3 models (PDB: 9Z6Y, 9Z6Z) and the accompanying pan-bacterial phylogenomic dataset make that question experimentally accessible, and comparative structural and phylogenetic analysis across the UG-RT family is, in my assessment, where the most informative next results will be found.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. Gemini (Google AI 3.1) was used for language polishing and structuring the perspective analysis.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Carla Ribeiro Polycarpo, Federal University of Rio de Janeiro, Brazil

Reviewed by: David McMillen, University of Toronto Mississauga, Canada