Diversity and genome engineering applications of R2 non-LTR retrotransposons Authors: Nozhat T. Hassan, Anna E. Sheppard, David L. Adelson, Kathleen Collins Published: 2026 ABSTRACT The R2 retrotransposon family has been a leading model system for studies of site-specific non-long-terminal-repeat retrotransposon mobility. More recently, avian R2 systems have been harnessed to implement stable sequence supplementation of the human genome. Investigations of R2 retrotransposons range from detailed biochemical studies of purified components to elucidation of the cellular processes that support new retrotransposon insertions. In addition, molecular and genetic studies have examined the persistence and diversification of R2 retrotransposons across animals. In publications starting from early 2024, an R2-encoded protein has been used in native form, or after modification, to insert transgene sequences of choice into heterologous genomes lacking an endogenous R2. Many of these genome engineering studies were launched using a historically accumulated reporting of R2 sequences informative for understanding retrotransposon evolution but not intended as a catalog of active R2 retrotransposons. Other groups approached R2 sequence identification by using or building consensus retrotransposon sequences from mostly short-read genome sequencing or genome assemblies. Here, we describe R2 sequence discovery, phylogeny, and annotation, and we consider their influence on R2 protein applications including gene addition to the human genome. INTRODUCTION Transposable elements (TEs) are genetic sequences that can spread to populate a significant part of eukaryotic genomes (Han 2010; Wells and Feschotte 2020). Accurate classification and subsequent annotation are foundational for understanding TE molecular evolution and context within the host genome (Sotero-Caio et al. 2017; Carey et al. 2021). The process of TE annotation from genomic sequences varies across studies, introducing uncertainty in comparisons across the field. Broadly, researchers often use automated pipelines to characterize TEs, sometimes in combination with a degree of manual curation. This results in classification into families with unique names based on sequence characteristics and phylogenetic relationships, among other delimiters (Bao and Eddy 2002; Wicker et al. 2007). However, there is a risk of misclassification of TEs using this approach due to error propagation of misannotations (Bao and Eddy 2002; Hoen et al. 2015; Carey et al. 2021; Hassan and Adelson 2023). For functional studies, some researchers compile sequences to build a TE consensus while others choose a representative sequence for further analysis. This choice can have a major impact on experimental findings. Recently, significant effort has been invested in expanded identification (Fig. 1A–C) and functional profiling of R2 retrotransposons (Chen et al. 2024; Edmonds et al. 2025; Fell et al. 2025; Zhang et al. 2025b; Hassan et al. 2026). R2s are a family of non-long terminal repeat (non-LTR) retrotransposons that insert site-specifically into sequences encoding 28S ribosomal RNA (rRNA), which are transcribed as part of a precursor transcript from repeating units termed rDNA loci (Fig. 2A,B; Eickbush and Eickbush 2015). The single R2-encoded protein (R2p) recognizes and nicks a 28S rDNA target site and uses the nicked DNA as a primer for cDNA synthesis, a mechanism termed target-primed reverse transcription (TPRT) (Fig. 2C; Luan et al. 1993). Many R2 structural features contribute to its continued mobility (Eickbush and Eickbush 2015; Deng et al. 2023; Wilkinson et al. 2023; Thawani et al. 2025). For example, 5′ and 3′ untranslated regions (UTRs) flanking the ORF (Fig. 2B) are essential to the retrotransposon life cycle. Key to applications described below, the 3′-UTR is where R2p binds and positions the R2 RNA 3′ end for TPRT (Luan and Eickbush 1995, 1996). In R2p, RNA and DNA binding surfaces span the full length of the protein, including in the central reverse transcriptase (RT) domain (Deng et al. 2023; Wilkinson et al. 2023; Thawani et al. 2025). The C-terminal region contains a CCHC zinc-knuckle motif followed by a “restriction-like” endonuclease domain with structural homology with Holliday-junction resolvases (Govindaraju et al. 2016; Deng et al. 2023; Wilkinson et al. 2023; Thawani et al. 2025). The N-terminal region can contain 1–3 zinc-finger domains (ZnF), which were historically used to categorize R2s into clades A (3 ZnF), B and C (different combinations of 2 ZnF) and D (1 ZnF) (Kojima and Fujiwara 2005). Alternatively, instead of domain configuration, sequence comparisons powered by a phylogenetically broadened sampling of R2s grouped R2p into either lineage A, with 3 ZnF, or lineage D, with predominantly but not exclusively 1 ZnF (Fig. 1C; Hassan et al. 2026). Lineage D R2p are the most widespread and anciently diverged into independently propagating sub-lineages. In comparison, lineage A R2p appear monophyletic, are detectable in extant animal genomes after divergence of cnidarians, and have a unifying co-fold of the most N-terminal two ZnF into an extra anchor for RNA and DNA (Thawani et al. 2025; Hassan et al. 2026). Summary of R2 retrotransposon distribution and N-terminus domain architecture. (A) Major phyla that contain species with R2s are on the inner tree. Bars represent the number of R2s in the combination of published studies and sequences deposited in Repbase (Kapitonov et al. 2009; Kojima et al. 2016; Hassan et al. 2026). For each phylum, icons of example species and their name are shown. The number of R2s in each phylum is as follows: Porifera—9, Ctenophora—4, Echinodermata—14, Cnidaria—15, Mollusca—10, Priapulida—1, Nematoda—1, Hemichordata—1, Arthropoda—72, Tunicata—9, Platyhelminthes—8, Annelida—5, Chordata—308. (B) Summary of R2s annotated in Chordata, with icons of example species in subgroups. The size of the bubble corresponds to the number of R2s annotated. (C) Schematic of the N-terminus domain architectures of R2p from lineage A and D. ZnF numbering increases toward the R2p N-terminus, rather than toward the C-terminus, due to designation of the universal ZnF as ZnF1. Lineage A R2p have co-folded ZnF3 and ZnF2 (“3:2”) in addition to ZnF1. Lineage D R2p typically have the single ZnF1 and a single Myb domain, but additional domain combinations have been inventoried as illustrated. rDNA unit features and steps of R2p-mediated gene insertion. (A) Schematic of Mammalia and Reptilia rDNA units with key features such as 18S, 5.8S, and 28S rRNA coding regions, Internal transcribed spacer 1 and 2 (ITS1-2), and Intergenic spacer (IGS) (Macdonald et al. 2024). The Reptilia rDNA unit is shown both uninterrupted and with an R2 insertion in the 28S rRNA sequence. (B) An expanded view of a Reptilia R2-inserted rDNA unit. R2 ORF is flanked by 5′ and 3′-UTRs. (C) Key steps of R2-mediated gene insertion: (1) R2p with bound template RNA locates the 28S rDNA insertion site, (2) R2p endonuclease (EN) domain creates a nick on the bottom strand, (3) R2p reverse transcriptase (RT) domain uses nicked DNA as a primer for cDNA synthesis, (4) R2p nicks the top strand, (5) second-strand DNA synthesis occurs. EN and RT domains are depicted as discrete modules when useful for visual clarity. R2s are increasingly of interest for engineering an RT enzyme to achieve gene delivery into eukaryotic genomes. Gene knockout and editing applications of Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR)/Cas systems have reinvigorated efforts for genome manipulation as therapy (Li et al. 2023; Kliegman et al. 2024; Liu et al. 2025; Zhang et al. 2025a). As a complementary approach, non-LTR retrotransposon proteins that use RNA as a template to synthesize entire genes into a genome are being developed as a tool for genome supplementation. Several recent studies utilize an avian R2p for insertion of short or multikilobase sequences into the human genome (Chen et al. 2024; Lee et al. 2024; Palm et al. 2024a; Edmonds et al. 2025; Fell et al. 2025; McIntyre et al. 2025; Thawani et al. 2025; Zhang et al. 2025b). Precise RNA-mediated insertion of transgenes (PRINT [Zhang et al. 2025b]) and site-specific target-primed insertion via targeted CRISPR homing of retroelements (STITCHR [Fell et al. 2025]) are two such technologies that employ an avian R2p for gene insertion. The latter, as the name suggests, fuses R2p with Cas9 as part of retargeting gene insertion to loci beyond the native R2 rDNA target site, directed by guide RNAs. These and other efforts, described in more detail below, used public or private sources of R2 sequence data obtained in previous studies of R2 evolution (Chen et al. 2024; Edmonds et al. 2025; Zhang et al. 2025b), or a custom database of non-LTR retrotransposon sequences mined from genome assemblies (Fell et al. 2025), to screen for activities of interest using biochemical and/or cellular assays. Each study used a unique approach for generating R2 sequences from different species and/or choosing the N-terminal position of exogenously expressed recombinant R2p. Here, we examine how R2s have been inventoried and annotated and how these processes influenced different groups’ efforts to adapt R2 systems to mediate new sequence insertion into a genome. Challenges in the annotation of R2 retrotransposons Since the initial discovery of R2 in Drosophila melanogaster rDNA (Burke et al. 1987), many studies have expanded our knowledge of R2 distribution across animal phylogeny. Historically, there has been a bias toward R2 discovery and investigation in arthropods, especially in insect genomes (Kojima et al. 2016; Bonandin et al. 2017; Zagoskina et al. 2020). Prior to the recent surge in R2 research, systematic analysis of R2 distribution identified a monophyly of R2s in Aves (birds), two clusters in Actinopterygii (ray-finned fishes), and the first R2s detected in Ctenophora (marine invertebrates), Hemichordata (worm-like organisms) and Mollusca (Kojima et al. 2016). R2s present unique challenges in annotation due to the presence of multiple distinct retrotransposon copies in any one organism and, if the R2 is mobile, across individuals of the same species. Also, some species harbor more than one independently perpetuated version of R2. In addition, many R2 insertions are 5′-truncated, which is an inherent outcome of cDNA synthesis initiation at the R2 RNA 3′ end (Perez-Gonzalez and Eickbush 2002; Han 2010; Levin et al. 2025). An R2 consensus assembled from short-read sequences will reflect contributions from truncated as well as full-length R2 insertions, incorporating both nonfunctional and functional retrotransposon copies. Therefore, consensus sequences will often encode inactive rather than active R2. As another compounding factor, rDNA is often underrepresented in genome assemblies due to repeat masking or technical barriers in resolving tandem repeats, and therefore rDNA-embedded R2s are also underrepresented. Furthermore, because R2p is translated by an unknown, noncanonical mechanism that does not oblige a methionine start codon (George and Eickbush 1999), there is not a deducible position of translation initiation; indeed, recent studies of recombinant R2p encoded by the same ORF used N-terminus positions widely spread across a region of ∼250 amino acids preceding the most N-terminal ZnF (see Supporting Information 1). Incomplete and error-containing R2 annotations impose limitations on the utility of sequence comparisons. Repbase is a commonly used database for TEs and other repeats in eukaryotic genomes, with access by registration for academic researchers (Jurka et al. 2005). Repbase preserves unique, valuable TE inventories, but Repbase R2 entries should not be treated as a collection of active retrotransposons; instead, these are sequences that can be incomplete and contain ambiguous base calls or errors (Hassan et al. 2026). Versions of what is biologically the same TE reported in Repbase may be accessible in GenBank, discoverable by separate searches of each relevant database (e.g., genome assemblies, genome sequencing reads, or publicly deposited sequence records). GenBank contains many more TE sequences than Repbase, but they frequently lack annotation. Repbase entries could cross-identify GenBank records to provide knowledge of more recent sequencing that is more accurate and in many cases more complete. Because protein homology searches have often been used for TE identification, resulting in initial reporting of only the TE ORF, added annotation that links this kind of record to its 5′ and 3′-UTRs and flanking target-site sequences would be valuable. Comparison of recent approaches for R2p use for genome engineering It is likely that all mammalian genomes lack R2 (Kojima and Fujiwara 2004; Kojima et al. 2016; Hassan et al. 2026). Nonetheless, expression of a heterologous R2p in mammalian cells supports RNA-templated insertion of new genes (see Table 1 for a summary and comparison of recent approaches) (Chen et al. 2024; Edmonds et al. 2025; Fell et al. 2025; Zhang et al. 2025b). The first published method, PRINT, utilizes codelivery of an mRNA encoding a native R2p and a template RNA encoding a payload sequence of choice to install autonomous transgene expression cassettes in mammalian cell rDNA at the highly conserved R2 target site (Palm et al. 2024b; Zhang et al. 2025b). PRINT was developed by screening recombinant R2p from organisms known to harbor R2, after an R2 inventory expansion in 2016 (Kojima et al. 2016). Protein sequences were designed from the combination of R2 annotations and BLAST searches of an annotated R2p against genome sequencing reads. Initial screening assayed purified R2p for highly selective use of an R2 3′-UTR RNA as template for TPRT (Zhang et al. 2025b). Avian R2p were found to have the highest specificity of template use. Cellular assays then compared R2p from Taeniopygia guttata (zebra finch TaGu; the same protein referred to as R2-1_Tg or R2Tg in subsequent studies described below), Zonotrichia albicollis (white-throated sparrow ZoAl), Geospiza fortis (medium ground finch GeFo) and Tinamus guttatus (white-throated tinamou) for rDNA insertion of a GFP reporter cassette. R2 3′-UTRs from these species were tested in cross-combination with each R2p, identifying an optimal heterologous pairing of TaGu or ZoAl with the GeFo 3′-UTR. The rationale for optimal PRINT with GeFo 3′-UTR became evident later: the avian R2 3′-UTR structure important for binding R2p (Thawani et al. 2025) is well-folded using the GeFo 3′-UTR sequence. Avian R2p use of R2 3′-UTR RNA as a TPRT template required 3′-UTR extension by a 3′ tail with at least 4 nt of rRNA sequence that can base-pair with the target-site primer 3′ end, and longer rRNA lengths could be counterproductive (Zhang et al. 2025b). Template RNAs that used an optimal transgene promoter sequence and contained pseudouridine or methylpseudouridine rather than uridine increased the percentage of full-length gene insertions. Cells with up to approximately five transgenes maintained transgene reporter expression without growth or viability compromise over many months in culture. TABLE 1.Summary of recent studies using an R2p for sequence addition to the human genomeStudySequence sourcesInitial R2 selection criteriaStudy goalsShortcomingsaPRINT (Zhang et al. 2025b)Repbase, genome short readsBiochemical analysis of R2p TPRT activity and template selectivity. Cellular 2-RNA delivery assays for transgene insertion.RNA-only delivery of R2p mRNA and template RNA for transgene insertion to the R2 target site.Use of biochemical assays as a filter has the risk of failing to predict cellular outcomes. Chen et al. 2024 RepbaseCellular assay of delivered plasmids’ ability to generate R2 target-site insertions. Proof of concept with all-RNA delivery.Plasmid-based expression of R2p and template RNA for transgene insertion to the R2 target site.Reliance on arthropod-biased R2 database. Optimization without R2p RT-dead control to monitor plasmid-templated insertions. Edmonds et al. 2025 RepbaseCellular assay of delivered R2 RNA ability to generate R2 target-site insertions. Subsequent 2-RNA delivery assays for transgene insertion.RNA-only delivery of R2p mRNA and template RNA for transgene insertion to the R2 target site.Reliance on arthropod-biased R2 database.STITCHR (Fell et al. 2025)Genome assembliesComputational pipeline to discover non-LTR retrotransposons with an R2-type endonuclease domain. Cellular assays of delivered plasmids’ ability to support sequence insertions to plasmid and rDNA or other genomic regions.Programmable, scarless sequence insertion to genomic loci of choice.Reliance on plasmid expression of protein, and on plasmid expression of template RNA for gene-length insertions. Insertions evident using R2p RT-dead controls, indicative of plasmid-templated genome insertions. R2Tocc failed rDNA insertion likely due to incorrect automated 3′-UTR annotation.aNot considering analysis of target-site specificity. Summary of recent studies using an R2p for sequence addition to the human genome aNot considering analysis of target-site specificity. A second approach used co-transfected plasmids and subsequently RNAs to install new sequences into mammalian cell rDNA (Chen et al. 2024). Almost all experiments relied on plasmid expression of template RNA, which changes R2p ribonucleoprotein assembly requirements relative to PRINT because template RNA expressed from a plasmid would be less susceptible to delivery-associated RNA hydrolysis and does not require nuclear import. Repbase R2p sequences were screened for potential full-length proteins, resulting in a list of 62, including different sequences from the same organism. Sixteen R2p, about half from Drosophila species that have been model systems for molecular characterization of R2 (George and Eickbush 1999; Perez-Gonzalez and Eickbush 2002; Nelson et al. 2023), were tested in human cells, paired with a template RNA encoding a GFP reporter cassette flanked by native R2 5′ and 3′ regions and 100 nt of rRNA sequence on each side. Experimental designs using plasmid-encoded template RNA with 100 nt homology tracts add two critical considerations: this allows plasmid-templated gene insertions, stimulated by R2p endonuclease activity at the target site, and it can lead to false-positive PCR products detecting transgene junctions due to cross-priming of separate linear amplification products from the genome and plasmid (Wang and Wang 1996; Potapov and Ong 2017). From the plasmid-system screen, R2-1_Tg (TaGu above, R2Tg below) was selected as the starting point for improvement by mutagenesis and other engineering, ultimately generating TgOPT. TgOPT increased the number of genome insertions relative to R2-1_Tg (Chen et al. 2024), leading to the conclusion that protein engineering was critically important for efficient R2p function in human cells. As a general statement, this conclusion is at odds with the successful use of native avian R2p for transgene insertions by several other groups (Edmonds et al. 2025; Zhang et al. 2025b; Ali et al. 2026; Muchenje et al. 2026). Also, one of the R2-1_Tg regions predicted to be disordered by Chen et al. and used successfully for heterologous domain insertions in plasmid-based R2p optimization was later found to a highly ordered loop that plays a role critical for PRINT with R2p expression from mRNA (Thawani et al. 2025). Clearly, much remains to be understood about how features of the same R2p can be differentially limiting in different expression contexts. A more recent study by Edmonds et al. replicated PRINT and extended it by delivery optimization and RNA chemical engineering (Edmonds et al. 2025). The authors expressed R2Tg using the PRINT mRNA design with BioNTech COVID-19 vaccine 5′ and 3′-UTRs and segmented polyadenosine tract (see Supporting Information 2). The transgene payload of template RNA was also largely the same, with the same uniquely optimal PRINT transgene promoter and a minimal polyadenylation signal. Initial screening used native-like retrotransposon RNAs to investigate human genome insertion activity of eight R2s of unstated origin, likely obtained from Repbase, including the unnoted duplicate of R2Sm-A and Perere-9, which are the same retrotransposon from the blood fluke Schistosoma mansoni (Kojima et al. 2006). Of these, R2Tg had the highest activity. Switching to the 2-RNA system, Edmonds et al. found that the template RNA features optimal for PRINT in human RPE-1 cells (Zhang et al. 2025b) were also optimal in HEK293FT cells, with small differences. The authors then enhanced transgene insertion efficiency by chemical engineering of the template RNA and improved lipid nanoparticle formulation, which are beyond the scope of this perspective to discuss. In future studies, optimization potentially building from the improvements demonstrated by Edmonds et al. and/or from other efforts will support translation of this technology to genomic medicine. The premise of STITCHR is to exploit the RT activity of a retrotransposon protein for scarless sequence insertions at genome loci of choice, targeted by RT-protein fusion to Cas9 nickase co-expressed with guide RNAs that direct nicking of both target-site strands (Fell et al. 2025). Screening and optimization relied entirely on plasmid-based expression of proteins and almost entirely on plasmid-based expression of template RNA. Fell et al. first generated an inventory of candidate site-specific non-LTR retrotransposons from genome assemblies using hidden Markov models for RT and restriction-like endonuclease domains. Many of the non-LTR retrotransposons in this collection are NesL/Utopia, Hero, and CRE elements (Malik et al. 1999; Kojima and Jurka 2015), in addition to R2s. The vast majority of sequences were from arthropods. When examining this data set, with the caveat that the published supplementary information record does not include database submissions, Drosophila species were the most highly represented. Historically, very few retrotransposons classified as R2 lack canonical R2 target-site specificity (Kojima et al. 2016). However, in the Fell et al. survey, R2Tg was reported to have upstream 28S rDNA sequence but no downstream rDNA. This conclusion is contrary to the fact that R2Tg has long been known to be embedded at the canonical R2 28S rDNA target site in genome sequences (Kojima et al. 2016), and that previously published functional studies recapitulated this rDNA insertion specificity in mammalian cells (Chen et al. 2024; Zhang et al. 2025b). R2Tocc, an initially mis-assigned avian R2 discussed in more detail below, was annotated to have upstream 28S rDNA and downstream 18S rDNA. This order is opposite the relative orientation of coding regions for 18S and 28S rRNA in an rDNA repeat, and it does not match the organism's genome assembly in which R2 is embedded at the canonical 28S rDNA target site (see Supporting Information 3). Fell et al. first tested 12 non-LTR retrotransposon systems, including 10 R2s, for the ability of plasmid-expressed protein and plasmid-expressed template RNA to support exon insertion into an R2 target site on a co-transfected reporter plasmid in HEK293FT cells (Fell et al. 2025). The template RNA included an R2 3′-UTR matched to the retrotransposon protein, but for these experiments whether flanking target-site homology was included was not specified. R2Tg was the most successful in sequence insertion to the plasmid. Curiously, removing the R2 3′-UTR binding site for R2Tg from the template RNA diminished insertions only by half, despite expected elimination of TPRT. For STITCHR retargeting of insertions, R2Tocc was selected as the R2p of choice for at least two reasons. First, when assayed by Fell et al., it was not active for 28S rDNA insertion. This failure likely resulted from incorrect annotation of the native R2 3′-UTR: it was 3′-truncated and therefore would be inactive as a template RNA (see Supporting Information 3). Second, R2Tocc out-performed R2Tg for retargeted insertion when fused to Cas9 nickase and co-expressed with two guide RNAs and a template RNA that harbored the insertion payload flanked by long tracts of homology with the intended target site. Because the goal was scarless insertion, the template RNA lacked the R2 3′-UTR binding site for R2p. The sequence record from which R2Tocc was derived was later flagged as a contaminant of Talpa occidentalis (Iberian mole) genome sequencing and is instead from Serinus canaria (Atlantic canary; see Supporting Information 3), an R2 embedded in 28S rDNA that had been described prior to the STITCHR survey (Kojima et al. 2016). Sequence contamination and subsequent retraction have become commonplace as genomic data are being produced at an increasing pace (https://www.ncbi.nlm.nih.gov/genbank/statistics/). This trend increases the importance of quality control in TE annotation, including TE taxonomic validation and verification of automated target-site predictions, especially when these properties influence major conclusions. Expanded R2 diversity and lineage classification We recently sought to expand the known distribution of R2, focusing on animal phyla previously under-inventoried for this retrotransposon (Hassan et al. 2026). Broader knowledge of R2 diversity will enable structure-function studies from an evolutionary perspective as well as providing additional insight into R2 phylogeny. Our approach was to implement an annotation method that retrieved full-length, rDNA-embedded, potentially functional R2s. This annotation method can be summarized in two main steps: (1) we used long reads where possible to extract full-length R2s, and (2) we had manual checks for 28S rDNA flanks, UTR boundaries, and identification of protein domains. Manual curation could be considered a drawback for implementation, but it was critical for correctly identifying and annotating divergent protein domains (e.g., different cysteine spacings in ZnFs) and validating precise UTR boundaries with rDNA. We did not attempt to evaluate R2 copy number per genome, which can be a useful indicator of recent R2 proliferation (Perez-Gonzalez and Eickbush 2002; Nelson et al. 2023). We chose the longest R2 ORF from any given species as the annotated R2p, or when more than one R2 lineage was evident in a species, we selected two sequences, one from each lineage. We note that annotation of the longest possible ORF does not indicate the actual R2p N-terminus, since R2p are noncanonically translated without a required methionine start codon (George and Eickbush 1999). As described above, this unusual feature led to the use of different N-terminal boundaries for the same R2p used for genome engineering by different research groups (see Supporting Information 1). Our recent study revealed that R2p has more domain variation than anticipated, and that the classification of R2s into clades based solely on the number of N-terminal ZnF is not a reliable reflection of their phylogenetic origin (Hassan et al. 2026). Expanding from previous analyses (Kojima and Fujiwara 2005; Kojima et al. 2016), we suggest that R2s have propagated as two major successful lineages, A and D, that are variably present in all major phyla from Ctenophores to Chordates (Fig. 1A; Hassan et al. 2026). Although Chordates currently have the highest number of species with annotated R2s (Fig. 1B), R2 remains undetected in mammals. We speculate that R2 loss in mammals is linked to rDNA evolution, which gave all mammalian rDNA units a much longer intergenic spacer between the transcribed region of each rDNA repeat than is present in any other phylogenetic group (Macdonald et al. 2024). Protein folding predictions revealed unexpected structural homology between the co-folded ZnF3:2 of lineage A R2p and a pair of tandem, differently positioned ZnF in a subset of Porifera Utopia non-LTR retrotransposon proteins we identified in parallel (Hassan et al. 2026), which lack the hallmark R2 Myb domain. Utopias in some species have site-specificity for insertion into U2 small nuclear RNA genes, whereas in other species they are less target-site specific (Kojima and Jurka 2015). The target sites of Porifera Utopias were not readily assigned but were not 28S rDNA, and the Myb-domain-containing Porifera R2-homologous retrotransposons also did not appear targeted to rDNA (Hassan et al. 2026). We suggest that further mining of Porifera and phylogenetically proximal Ctenophora non-LTR retrotransposons could bring additional perspective on the origin of R2 adoption of 28S rDNA as a target site and the origin of R2 lineage A ZnF3:2. Concluding thoughts about R2 and future directions for research R2 retrotransposons have persisted for a remarkably long time, predominantly maintaining site-specificity in 28S rDNA across diverse animal lineages despite the fact that gene conversion and repeat loss in tandem rDNA units should regularly purge them from rDNA arrays (Hall et al. 2022; Hori et al. 2023). Although R2 durability is evident in its distribution across organisms, we lack a framework for understanding this durability. Knowledge of R2 copy-number dynamics across across individuals in a population or across generations is limited, and how R2 copy number per genome relates to evolutionary persistence is unexplored. Also, although studies of R2 in Drosophila species demonstrated retrotransposon mobility both in germline and somatic cells (Eickbush and Eickbush 2011), the generality of this relaxed regulation of transposon mobility is unknown. Of particular importance, in all but a single model system, we lack experimental insight into how R2 can positively or negatively influence organism viability and reproduction. Recent studies with Drosophila melanogaster R2 suggest that it functions to normalize rDNA copy number in the male germline, inferred from defective rDNA maintenance upon R2 depletion (Nelson et al. 2023). However, features of R2 in Drosophila suggest that a gain-of-function supporting R2 persistence could be specific to this group of organisms. In particular, Drosophila R2 RNA 5′ ribozyme self-cleavage of template RNA from the RNA Polymerase I rRNA precursor removes all upstream rRNA sequences, instead of retaining the typical ∼28 nt that become a cDNA 3′ end capable of annealing to the upstream target-site DNA (Eickbush et al. 2013). Because cDNA base-pairing to upstream target-site is the favored mechanism for priming second-strand synthesis (Palm et al. 2024a; McIntyre et al. 2025), lack of this base-pairing could stabilize a DNA break and favor homologous recombination (Nelson et al. 2023). Insertion into 28S rDNA is a hallmark of R2s, but in a few species, R2 deviated from this conserved targeting. Inventories of historical TE insertions, inferred from genome sequencing, suggest that in a few species R2 became a generalist lacking site-specific insertion, and in a few others, it switched to a new target site elsewhere in rDNA (Kojima et al. 2006, 2016; Gladyshev and Arkhipova 2009). Because there could be different rates of R2 loss from the general genome compared to rDNA, profiling insertion-site preference from extant genome locations may be misleading. Binary classification of new genome insertions as “site-specific” or not is likely an oversimplification of what could be a wide range of more or less stringent target-site specificity. Insertion-site specificity would be best evaluated experimentally, with minimal opportunity for viability or growth selection between the insertion process and insertion-site readout. Parallel assays of insertion-site preference using the same cell line or organism would be optimal for specificity comparison across different species’ R2, with the caveat that retrotransposon properties may differ in the often-unknown native physiological context for mobility. This effort could inform additional development of R2-based genome engineering. The feasibility of using a very species-divergent R2p to accomplish safe-harbor rDNA transgene insertion is highlighted by recent studies demonstrating avian R2p-mediated insertion of full-length transgenes into the rDNA of the plants Nicotiana benthamiana, Arabidopsis thaliana, Solanum lycopersicum, and Oryza sativa (Ali et al. 2026; Muchenje et al. 2026). Finally, we believe that field-wide consideration and implementation of best-practice principles would increase the impact of future non-LTR retrotransposon discovery, annotation, characterization, and deployment. At a first level, there could be consistent standards for validation, annotation, and experimental reporting of full-length or partial TEs, their ORFs, and their flanking target-site sequences. The providence of newly reported TE sequences, and of any TE sequences used in research assays, could be reported explicitly as sequence of an individual cloned retrotransposon or a sequence consensus, and as deriving from short- and/or long-read sequencing with a stated error correction protocol. Each publication should have sequence files that are searchable and keyed to Figures in which the sequences were used. At a second level, for profiling insertion-site specificity, the use of consistent principles and fully reproducible wet-bench and bioinformatic workflows would increase the transparency, validity, and utility of claims that without best-practice standards are based on disparately chosen definitions of “off-target” background. At a third level, a fully open-access community resource that unified databases and homogenized annotations with tracked updates of new sequence versions would be especially valuable for accelerating future research progress. SUPPLEMENTAL MATERIAL Supplemental material is available for this article. COMPETING INTEREST STATEMENT K.C. is a cofounder, board member, and consultant to Addition Therapeutics, Inc., reviewed and approved by the University of California, Berkeley in accordance with its conflict-of-interest policies; both she and the company may benefit from commercialization of the PRINT platform.