Mechanism-Guided Bioinformatic Analysis of Non-LTR Retrotransposon Insertions: From Molecular Archaeology to De Novo Experimental Systems

Abstract

Non-long-terminal-repeat (non-LTR) retrotransposons insert into the genome by priming cDNA synthesis directly from the target site in a process termed target-primed reverse transcription, but the subsequent steps that complete insertion remain poorly understood. In contrast to the relatively uniform products of integrase-mediated LTR retrotransposon integration, non-LTR retrotransposon insertions are structurally heterogeneous, complicating their analysis. Yet this heterogeneity can reveal variation in the underlying insertion reaction when interpreted in an appropriate mechanistic context. Here, we present a framework for mechanism-guided bioinformatic analysis of non-LTR retrotransposon insertions, based on the premise that distinct features of insertion architecture and genomic context report on different biochemical steps of the insertion process. Drawing on historical and recent studies, we identify best practices for sequencing and bioinformatic analysis and consider how extant genomic insertions and experimentally generated de novo insertions provide complementary mechanistic information while imposing different inferential biases. We envision that standardized principles for mechanism-guided analysis will help resolve poorly understood steps of insertion completion, enable comparative investigation across retrotransposons and hosts, and improve evaluation of emerging retrotransposon-based genome-engineering systems.

1. Introduction

Non-long-terminal-repeat (non-LTR) retrotransposons, sometimes called LINEs (long interspersed elements), are a class of eukaryotic transposable elements (TEs) that mobilize through an RNA intermediate. They are among the most ancient extant eukaryotic retrotransposons [1,2] and have achieved widespread evolutionary success, especially in animal genomes, where their activity in extreme cases has generated the majority of genome sequence [3,4].

Non-LTR retrotransposons predominantly initiate insertions through a mechanism called target-primed reverse transcription (TPRT), in which the retrotransposon-encoded catalytic-protein endonuclease (EN) domain nicks one strand of the target site, generating a 3′ hydroxyl that is used to prime reverse transcription of the bound template RNA (Figure 1A, EN-dependent insertion) [5]. Completing an insertion requires nicking the second strand of the target site, forming a 5′ junction, and synthesizing second-strand cDNA. Because the 3′ junction is established during TPRT initiation, whereas the 5′ junction is generated through subsequent and more variable completion steps, non-LTR retrotransposon insertions are structurally asymmetric and heterogeneous. This contrasts with the relatively uniform products of integrase-mediated insertion by long terminal repeat (LTR) retrotransposons, which occurs only after synthesis of full-length double-stranded DNA [6].

Figure 1. Insertion mechanisms and taxonomic diversity of non-LTR retrotransposons. (A) Schematics of endonuclease (EN)-dependent and EN-independent mechanisms of insertion by non-LTR retrotransposon proteins. “DSB” indicates a double-strand break. (B) Unrooted schematic cladogram of major non-LTR retrotransposon lineages based primarily on RT domain phylogenies [2,7,8]. Branch lengths are arbitrary, and poorly resolved deep relationships are shown as polytomies (multiple descendant branches from a given node). Group II intron RTs are shown as a putative outgroup. In the middle and on the right, the autonomous and nonautonomous elements depicted are representative examples of the respective groups, as indicated. The schematic does not imply that the example nonautonomous element is mobilized by the example autonomous element connected by a dashed line, except for human L1 and Alu. The group nomenclature and clade groupings follow the Repbase classification [2]. The “R2” group includes R2, R4, Hero and NeSL clades. The “L1” group includes Proto1, L1, and Tx1 clades. The “RTE” group includes RTETP, Proto2, RTEX, and RTE clades. The “I” group includes Outcast, Ingi, Vingi, I, Nimb, Tad1, Loa, R1, and Jockey clades. The “CR1” group includes Rex1, CR1, Kiri, L2, L2A, L2B, Crack, and Daphne clades. “APE” indicates apurinic/apyrimidinic endonuclease-like domain, and “RLE” indicates restriction-like endonuclease domain. “DBD” indicates DNA-binding domains. “CTD” indicates C-terminal domain. “RNH” indicates RNase H domain. “TBR” indicates tail-binding region. “SIDE” indicates short internally deleted element. “28S” refers to sequence from the 28S ribosomal RNA gene. “RZ” indicates ribozyme.

Early observations of non-LTR retrotransposon insertions showed precise 3′ ends and variable, often truncated, 5′ ends [9,10,11]. Together with homology of the encoded protein to known reverse transcriptases (RTs), the truncation profile suggested a model in which reverse transcription was primed directly from chromosomal DNA at a characteristic position near the 3′ end of the template RNA, explaining the precise 3′ junction of insertions (see Figure 1A), but used more variable mechanisms to join cDNA to the target site, accounting for 5′ heterogeneity. The central prediction that reverse transcription could be primed directly from the chromosomal target site was experimentally validated in 1993 by Eickbush and colleagues [5], establishing the TPRT mechanism now known to be shared across non-LTR retrotransposons. The genomics-informed discovery of TPRT illustrates the central premise of this review: sequences inserted by non-LTR retrotransposon proteins can serve as molecular records from which transient steps of retrotransposition can be reconstructed.

In the context of renewed interest in studying this class of retroelements [12], including a flurry of innovative retrotransposon-based strategies for transgene insertion [13], here we review historical approaches to and outline best practices for mechanism-guided bioinformatic analysis of inserted sequences. We organize bioinformatic evidence according to the steps of the insertion pathway that are informed by sequence signatures, beginning with the reconstruction of the original target site prior to insertion and continuing to mechanisms of target-site selection, first-strand cDNA synthesis, completion of the insertion, and genome rearrangements. Highlighted examples from the literature primarily focus on animal non-LTR retrotransposons, where most mechanistic work has been performed.

Bioinformatic discovery and characterization of non-LTR retrotransposons themselves—including de novo element discovery, reference/consensus sequence construction, classification and phylogenetic analysis, and prediction of potentially active copies—require a distinct set of approaches and are outside the scope of this review [13,14,15,16,17]. Additionally, recent advances in the use of non-LTR retrotransposon proteins for genome engineering have been reviewed elsewhere [12,13]. We instead focus on the analysis of the insertion events generated by these elements.

2. Structural and Biochemical Diversity of Non-LTR Retrotransposons

Here, we briefly review the structural and biochemical diversity of non-LTR retrotransposons, focusing on experimental models relevant to the subsequent discussion. Although deep phylogenetic relationships remain unresolved, current classifications recognize eight broad groups of known non-LTR retrotransposons, excluding Penelope-like elements (Figure 1B) [1,2,7,8,18]. R2, CRE, and Dualen group elements generally encode a single open reading frame (ORF); L1 and Ambal group elements generally encode two; and RTE, I, and CR1 group elements are more variable [2].

The largest protein, which is encoded by ORF2 of a two-ORF element or by the sole ORF, contains reverse transcriptase (RT) and endonuclease (EN) domains. A third catalytic domain, encoding RNase H activity, is observed in some ORF2 proteins (ORF2p). The presence of other domains and the organization thereof differ substantially among elements even within the same taxonomic group. EN domains are categorized as either apurinic/apyrimidinic endonuclease-like (APE) or restriction-like endonuclease (RLE) according to sequence homology (the presence of either or both EN domains is indicated for each element family in Figure 1B). When present, ORF1p is generally thought to facilitate ribonucleoprotein (RNP) formation and RNA trafficking, but its precise functions remain poorly understood, in part due to substantial variation in sequence and domain architectures across non-LTR retrotransposon lineages [19,20].

Five historically important model systems to highlight are insect R1, animal R2, human LINE-1 (L1), Drosophila I factor, and zebrafish LINE-2 (ZfL2) (Figure 1B). Biochemical, cell culture, and organismal systems for investigating these retrotransposons anchor many of the historical developments in the field. These five representatives differ along major dimensions of non-LTR retrotransposon diversity, including EN domain type, number of ORFs, and insertion-site specificity.

Both R1 and R2 elements insert site-specifically into the sequence encoding the 28S ribosomal RNA in a larger, multicopy repeated gene unit (rDNA), whereas L1, I factor, and ZfL2 elements insert in a dispersed manner. Silk moth Bombyx mori R1 (R1Bm) became an early model when its characteristic 14-bp target-site duplication (TSD) was explained by offset in vitro cleavage of target-site DNA strands, generating 3′ overhangs [21]. B. mori R2 (R2Bm) enabled the foundational biochemical reconstitution of EN cleavage and TPRT [5]. More recently, avian R2 proteins (R2p), particularly zebra finch Taeniopygia guttata R2p, have become models for retrotransposition biochemistry and cell biology [22,23], as well as tools for genome engineering (as reviewed in [12,13]). L1, the sole active autonomous transposable element in humans, is the primary mammalian model, supported by abundant data on somatic and polymorphic insertions and tractable de novo cell culture assays. Here, polymorphic insertions refer to germline-inherited insertions present in some but not all individuals in a population. Drosophila melanogaster I factor provides a genetically inducible in vivo model of germline retrotransposition and associated genome rearrangements through I-R hybrid dysgenesis [24]. Finally, zebrafish Danio rerio L2 elements belong to the CR1 group and variably encode one or two ORFs. ZfL2-1 and ZfL2-2 both stringently recognize their cognate 3′ UTRs [25], and the more extensively studied ZfL2-2 has been shown to mobilize in heterologous cell culture systems [26,27], enabling analysis of host-dependent insertion outcomes [28,29].

In addition to mobilizing their own transcripts, non-LTR retrotransposon proteins can make genome insertions by copying RNAs from nonautonomous elements, including short interspersed elements (SINEs), which lack their own retrotransposition proteins and depend on autonomous elements for mobilization (Figure 1B). RNA sequence or structural features can mediate recognition by the retrotransposon protein. Insertions created by copying cellular mRNAs generate processed pseudogenes.

3. A Framework for Mechanism-Guided Insertion Analysis

We ground this review in the principle that genome sequence analysis is informative for deducing individual biochemical steps across the non-LTR retrotransposon insertion process. As diagrammed in Figure 2, different parts of the insertion or surrounding target site report on different steps of the insertion reaction. Throughout this review, 5′ and 3′ insertion ends are designated relative to the positive-sense RNA template, rather than the initial cDNA strand (Figure 2A).

Figure 2. Sequence features of non-LTR retrotransposon insertions and related mechanistic interpretations. (A) Schematic representing inference of pre-insertion target site and mobilized template RNA from observed insertion-containing allele. “TSD” indicates target-site duplication. (B–F) Schematics identifying sequence features of retrotransposon insertions, junctions, or reconstructed pre-insertion target sites and relevant mechanistic interpretations for each step. “DSB” indicates double-strand break. “NTA” indicates non-templated addition of nucleotides. “NHEJ” indicates non-homologous end-joining.

Our framework begins with inferring the inputs of the insertion reaction: (1) reconstructing the pre-insertion target site and (2) reconstructing the mobilized RNA and its source (Figure 2A). It continues with inferring the biochemical steps of the insertion reaction: (3) target-site selection and first-strand cleavage (Figure 2B), (4) DNA primer generation and initiation of reverse transcription (Figure 2C), (5) first-strand cDNA synthesis (Figure 2D) and (6) second-strand cleavage (Figure 2E), 5′ junction formation (Figure 2F), and second-strand synthesis. Finally, this framework will address potential off-pathway products of the insertion: (7) insertion-associated genome rearrangements and (8) abortive insertion and collateral genome damage.

In traversing each step of this framework, we will review what evidence can be gathered from the analysis of fixed genomic insertions and from experimentally generated de novo insertions. Here, a fixed genomic insertion refers to an inherited retrotransposon insertion present in essentially all individuals in a population (allele frequency ~1). Where appropriate, we will also consider what evidence is provided by analysis of somatic transposition events or polymorphic insertions. Other kinds of evidence—including in vitro biochemistry and quantifications of retrotransposition frequency—will not be covered by this review except when relevant for context of bioinformatic analyses.

4. Sources of Mechanistic Evidence

The mechanistic conclusions that can be drawn from insertion sequences depend on three linked choices: how insertions arose (extant genomic insertions vs. experimentally generated insertions), how those insertions are sequenced, and how the resulting reads are analyzed bioinformatically. Fixed, somatic, polymorphic, and experimentally generated de novo insertions differ in the biological histories they capture; sequencing methods differ in which events and sequence features they recover; and computational pipelines differ in which aspects of insertion architecture they preserve and interpret. Together, these factors determine both the information available for mechanistic inference and the biases that constrain it.

Section 4.1, Section 4.2 and Section 4.3 consider the distinct strengths and limitations of fixed genomic insertions, somatic insertions, polymorphic insertions, and experimentally generated de novo insertions.

Sequencing method further determines which insertion events and sequence features are recovered. Method choice should be guided by the mechanistic feature of interest, with quantitative recovery of independent insertion events prioritized, where possible, through the use of unique molecular identifiers and minimal PCR amplification. Size selection should be avoided when insertion length or architecture is itself an outcome of interest. Table 1 summarizes representative sequencing approaches, their principal strengths and biases, and example applications.

Table 1. Assays for sequencing non-LTR retrotransposon insertions. The lists of assays and applications for each category are intended to be representative but non-exhaustive. The “S/L” column identifies whether the method is typically used with short-read (S) and/or long-read (L) sequencing technologies. “PCR bias” refers to sampling and capture biases associated with exponential amplification by PCR. Relevant sources of bias include length bias, sequence composition bias, stochastic dropout and jackpotting, limited or no detection of complex rearrangements, and requirement for insertion to contain primer-binding site for detection. “Insertion sequence requirement” indicates that sequencing method requires insertions to contain a specific sequence and that the associated method is therefore only compatible with de novo insertion assays.

S/L	Assay Category	Representative Examples	Design	Strengths	Weaknesses
S, L	Whole-genome sequencing (WGS)	[23,30]	Sequence genomic DNA without TE-specific enrichment	Minimal TE-specific enrichment bias; captures genome rearrangements; long-read WGS can recover both the full insertion and flanking sequence	Low coverage of insertions for given cost/sequencing depth; underpowered detection of rare events
S	Hybridization capture	RC-seq [31]; mRC-seq [30]	Capture of gDNA with designed probe library	Can multiplex across TE families; enrichment step is amplification-free	Under-captures junctions and flanking sequences; probe design is complex; probe libraries are expensive
S	Fragmentation + linker PCR	Restriction digest: transposon-seq [32]; TIPseq [33,34]	Restriction digest → splinkerette/vectorette adapter ligation → PCR with TE + adapter primer pair	Sensitive; inexpensive; established for TE insertion mapping	Restriction-site bias; PCR bias; usually cannot capture both junctions
		Random shear: ATLAS-seq [35]; HAT-seq [36]	Sonication → adapter ligation → PCR with TE + adapter primer pair	Removes restriction-site bias; highly sensitive; random fragment ends can distinguish independent insertions	PCR bias; usually cannot capture both junctions; sheared end coordinates are not equivalent to true UMIs
		Transposase: TagMap [37]; TTISS [22,38]	Tn5-based tagmentation → PCR with TE + adapter primer pair	Simple and fast library prep	Tn5 insertion bias; PCR bias; usually cannot capture both junctions
S	Inverse PCR	[39,40,41]	Restriction digest → intramolecular ligation → insert-specific PCR	Can recover long insertions and flanking sequence	Technically challenging; restriction-site bias; PCR bias
S	Traditional amplicon sequencing	[42,43,44]	PCR with primers flanking insertion site OR one primer near target site and one primer within insert	High coverage of site-specific precise insertion junctions	Recovery of on-target insertions from site-specific elements only; PCR bias; often non-quantitative
S	Linear amplification-mediated PCR	LAM-PCR [45]	Multiple cycles of single-primer extension from insert-anchored primer → exponential PCR	Linear enrichment provides high sensitivity	PCR bias; usually cannot capture both junctions
L	Linear amplification by phage polymerase transcription	RIPPLE-seq [46]	in vitro transcription from promoter sequence in insertion → reverse transcription	PCR-free; can capture long insertions; linear enrichment provides sensitivity and reduces bias	Insertion sequence requirement; low input necessary to capture both junctions
L	Insertion cloning	[47,48,49,50]	Restriction digest → intramolecular ligation → bacterial selection for insertion sequence → clonal plasmid prep	Can recover long insertions and flanking sequence	Limited throughput; insertion sequence requirement; some sequences unstable in bacteria
L	Cas9-targeted, amplification-free long-read sequencing	nCATS [51,52]; AFIS-seq [53]	Cas9 cleavage in/near insert	PCR-free; can capture long insertions; native DNA retains methylation	Lower enrichment than amplification-based methods; high DNA input; guide/cut-site optimization required

Finally, bioinformatic approaches determine which molecular features can be reconstructed from sequence data. Section 4.4 and Table 2 summarize computational approaches according to their utility for mechanism-guided insertion analysis and provide representative applications.

Table 2. Computational approaches for analysis of non-LTR retrotransposon insertions. Lists of tools and applications for each category are intended to be representative but non-exhaustive.

Analysis Class	Representative Tools	Primary Purpose	Limitations for Mechanism-Guided Insertion Analysis	Representative Applications
Repeat annotation	RepeatMasker	Identify and annotate sequence derived from known TE families	Annotates sequence fragments rather than reconstructing an insertion event; general TE annotation, not specific to non-LTR elements	[54,55,56]
TE insertion discovery	MELT, RelocaTE2, PALMER, xTea, TLDR	Find reference/non-reference TE insertions and resolve insertion sites/structures from genomic reads	Primarily optimized for locating and annotating insertions; does not reconstruct mechanistically informative steps of insertion	[57,58,59,60]
Amplicon analysis	CRISPResso2, knock-knock, custom pipelines	Align reads around a defined locus and classify editing outcomes	Genome-editing-centric categories; does not reconstruct mechanistically informative steps of insertion	[22,43,44]
Mechanism-guided reconstruction	Custom pipelines and manual curation	Explicitly reconstruct target-site allele, template-derived sequence, both insertion junctions, rearrangements, etc.	Powerful but poorly standardized; results can depend on analysis choices, including read filtering, alignment, and event-calling thresholds	[38,61,62,63]

4.1. Fixed Genomic Insertions

Fixed non-LTR retrotransposon insertions can be thought of as genomic fossils, and their study as “molecular archaeology” [54]. As with physical fossils, genomic insertions surviving to the present day do not represent a complete record of past insertion events. Insertions are eliminated by neutral genetic drift or purifying selection [64], and mutation, deletion, and recombination obscure information over time. These problems are particularly acute for insertions into repetitive loci such as rDNA or microsatellites, where high recombination rates and concerted evolution are associated with rapid turnover of insertions [65,66,67]. Despite its limitations, analysis of fixed genomic insertions provides the only route to study some non-LTR retrotransposon systems for which biochemical reconstitution or de novo insertion experimental systems are intractable or have not yet been developed.

For analysis of fixed insertions, not all source data are equally informative. When available, haplotype-resolved long-read assemblies enable the most comprehensive analysis of TE insertions [68,69]. However, as non-LTR retrotransposons commonly target multicopy loci encoding conserved non-coding RNAs or microsatellites, telomeres, and other low-complexity sequences [70], analysis of even high-quality, “telomere-to-telomere” genome assemblies may be insufficient for comprehensive analysis of TE insertions. For example, even in the most updated D. melanogaster genome reference (dm6 at the time of this writing), rDNA arrays are not fully assembled [71]. Therefore, a thorough investigation of R1 and R2 insertions would require an analysis of source reads, regardless of assembly status. Analysis of TE insertions is best powered by long reads, which can provide a molecular record of individual insertion events [59,72].

4.2. Somatic and Polymorphic Insertions

Although less common among published studies, analysis of somatic transposition events and of polymorphic insertions can be a powerful tool for generating or testing mechanistic hypotheses. Somatic events and polymorphic insertions avoid some of the bioinformatic challenges associated with fossilized genomic insertions while still reporting on retrotransposition in its native organismal context. These sequences have undergone relatively little decay, and an insertion-less allele of the target site can often provide the pre-insertion target-site sequence.

The primary challenge for analysis of somatic and polymorphic insertions is one of scarcity. Available datasets generally contain fewer of these insertion events than fixed genomic copies or induced de novo insertions. Human tumor samples are an important exception because they are deeply sequenced and broadly sampled by several consortia, providing a wealth of somatic transposition data [73,74,75,76,77].

4.3. De Novo Insertions

De novo sequence insertion assays enable the capture of new genome insertions before evolutionary selection or sequence decay, better preserving the signatures of all the supporting cellular steps. However, as a caveat, these assays do not fully recapitulate the abundance, stoichiometry, localization, or processing of retrotransposon components that occur during endogenous retrotransposition.

We distinguish three primary classes of de novo insertion experiments that differ in how retrotransposition is initiated. The first class involves the delivery of DNA, often episomal, to express non-LTR retrotransposon RNA and proteins (Figure 3A). Traditional DNA-based expression assays, particularly for L1, most often contain an antisense reporter gene (typically conferring drug resistance or fluorescent protein expression) interrupted by a sense intron, thus allowing the selection of cells with bona fide retrotransposition events [78,79]. DNA-based expression constructs typically encode the retrotransposon machinery and template RNA in cis, but separate expression constructs for retrotransposon protein and template RNA have been used for studies of nonautonomous element mobilization and trans complementation [80,81,82,83]. Host-cell machinery determines the level and fate of RNA transcripts in a non-physiological expression context, with potential heterogeneity in the pool of molecules serving as templates for reverse transcription. For downstream sequencing applications, the presence of the donor sequence in a host chromosome or episome can overwhelm signal from RNA-templated genome insertions.

Figure 3. Experimental systems for generating de novo non-LTR retrotransposon insertions. (A) Schematic of LINE-1 retrotransposition assay involving delivery of episomal construct expressing L1 ORF1p and ORF2p and encoding a sense-orientation intron within an antisense neomycin resistance gene (“neo”). Upon insertion of the RNA encoding the L1 machinery and spliced neo expression cassette (in cis), previously susceptible cells (“G418^S”) become drug resistant (“G418^R”). “SD” and “SA” indicate splice donor and splice acceptor, respectively. Based on Moran et al. (1996) [78]. (B) Schematic of R2 retrotransposition assay involving delivery of R2p-encoding mRNA and transgene-encoding RNA. Upon insertion of the RNA, cells express GFP. “pro” indicates promoter and “pA” indicates polyadenylation signal sequence. Based on the Precise RNA-mediated INsertion of Transgenes (PRINT) system [23]. (C) Overview of I-R hybrid dysgenesis system for in vivo induction of I factor transposition in D. melanogaster. Schematic depicts genetic cross between female individual from a Reactive strain and male individual from an Inducer strain producing sterile F₁ females (“SF”) due to I factor transposition. “×” symbol indicates a lack of transposition events. “✓” symbol indicates active transposition.

The second class involves the delivery of RNA or RNP (Figure 3B). In a cis configuration, an mRNA encoding a non-LTR retrotransposon protein provides both a template for protein translation and a template for TPRT [22,44]. In a trans configuration, an mRNA encodes a retrotransposon protein, while a separate RNA serves as template for TPRT (Figure 3B) [22,23]. A trans format facilitates the separate investigation of the requirements for the RNA template and retrotransposon proteins for insertion. A less common format involves the delivery of a preformed RNP, providing maximal control over the molecular complex responsible for insertion [84]. RNA- and RNP-based modes of delivery restrict insertions to a defined pulse of activity in comparison to plasmid delivery, such that insertion outcomes can be read out within hours of delivery to reduce potentially confounding variables such as cell survival or host silencing of transgenes. Additionally, delivery of the template RNA as a purified transcript has several benefits: it can provide a specific sequence rather than a collection of heterogeneously processed transcripts, and the template RNA can use nucleotide analogs and chemical modifications to engineer a desired RNA fate or half-life.

A third class of de novo insertion experiments entails inducible mobilization of endogenous non-LTR retrotransposons in vivo (Figure 3C). I-R hybrid dysgenesis in D. melanogaster is a model system, in which controlled genetic crosses induce I factor mobilization in the germline of female progeny [85]. Assays relying on mobility of endogenous elements retain physiological context but lack an opportunity to readily alter insertion sequence and to constrain the tempo and molecular details of retrotransposition.

4.4. Bioinformatic Tools and Mechanism-Aware Insertion Analysis

We have evaluated bioinformatic tools for their capacity to analyze insertions in a mechanism-aware manner. Table 2 categorizes computational tools for (1) repeat annotation, (2) de novo, somatic, and polymorphic insertion discovery, (3) amplicon analysis, and (4) mechanism-guided insertion analysis, each with literature references.

Repeat annotation tools such as RepeatMasker identify TE-derived sequence in a host genome assembly using an input library of TE sequences [86]. These annotations are not tailored to features specific to non-LTR retrotransposon insertions but can support custom downstream analysis of truncations, inversions, and junction architectures [54,55]. Insertion-discovery tools—including MELT [58], TLDR [87], and PALMER [59]—infer de novo, somatic, and polymorphic insertion events by comparison of sequencing reads to a host genome assembly. Some insertion-discovery tools, including TLDR and PALMER, additionally reconstruct the insertion architecture, including internal features such as truncations, inversions, and transduced sequence (described below), as well as target-site alterations.

Beyond repeat annotation and insertion-discovery tools, a third category of software analyzes amplicons for genome insertion or editing outcomes. Recent studies adapting non-LTR retrotransposons into genome-engineering tools have often analyzed de novo insertions using genome editing software such as CRISPResso [22,43,44,88,89]. CRISPResso estimates relative insertion efficiency and identifies imprecise junctions or indels within a pool of sequenced amplicons, with the caveat of length-dependent PCR amplification bias; it does not reconstruct the mechanistically informative features of non-LTR insertions.

At the time of writing, there are no widely used bioinformatic tools broadly capable of reconstructing insertion events mediated by non-LTR retrotransposon machinery. However, numerous historical and recent publications provide examples of bespoke analyses that have been applied effectively to mechanistic inquiry [35,38,45,46,47,48,61,62].

5. Reconstructing Inputs, Outputs, and Reaction Steps

5.1. Inputs: Reconstructing the Pre-Insertion Target Site

Reconstruction of the pre-insertion target site (Figure 2A) is a prerequisite for inferring insertion-associated target-site alterations, but this process is challenging for fossilized genomic insertions. Unlike the characteristic 8-bp TSDs associated with the mobility of D. melanogaster P element DNA transposons [90], for example, non-LTR retrotransposon insertions can result in TSDs, target-site truncations, or genomic addition of extra nucleotides, with heterogeneity across insertion loci carrying the same retrotransposon in the same genome. Any attempt to reconstruct an uninserted target-site allele for fossilized insertions is inherently indirect, in that the accumulation of mutations after insertion obscures the ancestral state.

Strategies for target-site inference can be broadly classified as reference-agnostic or alignment-based. The primary reference-agnostic method can be applied only to the subset of insertions that generate TSDs, which can be collapsed to reconstruct the pre-insertion target site. However, even the subset of non-LTR retrotransposons that canonically form TSDs do not always generate TSDs at all insertion loci [54]. For example, when L1 makes EN-independent insertions instead of EN-dependent insertions (Figure 1A), TSDs are absent [28,48,55,91].

Another approach is to align the insertion-containing locus to reference sequence(s) for the same or phylogenetically proximal organisms to infer the pre-insertion target site [92]. This approach has been useful even for insertions in multicopy loci or low-complexity sequence repeats, as demonstrated, for example, for retrotransposons targeting ribosomal RNA genes, U2 small nuclear RNA genes, or other repetitive sequences [16,28,70,93,94,95,96,97,98,99]. However, analysis of insertions into these multicopy target sites poses a unique challenge, in that definitive identification of the specific chromosome position of insertion can be impossible. In these cases, alignment to a consensus reference sequence of a multicopy locus is more accurate, and more appropriate, to report than random coordinate assignment of multi-mapping reads. Anchoring to a consensus reference simplifies downstream analysis, for example in identifying patterns of target-site alterations [23].

In contrast to target-site reconstruction for preexisting mobility events, target sites for de novo insertions can usually be reconstructed by aligning insertion-flanking sequence to the reference genome. Complex rearrangements involving multiple sequence sources, inversions, deletions, or duplications may nevertheless remain difficult to detect and reconstruct [49]. Consequently, such events are often overlooked or not systematically evaluated in insertion sequence analysis. However, recent advances in long-read sequencing, bioinformatic methods, and reporter assays are beginning to enable their detection and mechanistic characterization (see Section 5.7 and Section 5.8).

5.2. Inputs: Reconstructing the Mobilized RNA and Its Source

A parallel effort to target-site reconstruction is inference of the template RNA that gave rise to the inserted sequence (Figure 2A). For endogenous retrotransposition, this can also help identify the source element. In native contexts or de novo insertion assays using plasmid-based expression of a template RNA, transcription and RNA processing can generate heterogeneous templates through variable transcriptional boundaries, splicing, polyadenylation, and self-cleavage. The inserted sequence can diverge further from the actual template RNA sequence due to internal initiation of reverse transcription, non-templated nucleotide additions, premature RT termination, and variable 5′-junction formation by host-cell machineries (Figure 2C,D). RNA ligation and template jumping can further alter the sequence that is ultimately inserted [100,101,102]. Mutations introduced during cDNA synthesis or accumulated after insertion add another source of divergence. Direct delivery of template RNA simplifies reconstruction by defining the input sequence, although heterogeneity can still arise during in vitro transcription or through RNA degradation after delivery [103,104,105].

Inefficient use of the L1-encoded polyadenylation signal can extend transcription into downstream genomic sequence, producing 3′ transductions when this sequence is subsequently inserted [106,107,108,109]. Imprecise transcription initiation or readthrough from upstream promoters can similarly generate 5′ transductions [110,111]. Conversely, inserted sequences do not necessarily reflect the entire transcript sequence: for example, D. melanogaster I factor transcripts include downstream flanking sequence that is not incorporated into insertions [112,113].

Reconstructing the mobilized RNA therefore requires considering alternative RNA-processing products rather than assuming that the inserted sequence derives directly from an annotated retrotransposon transcript. Cryptic splicing of L1 transcripts, for example, generates internal deletions detectable in both RNA and insertion products [114,115]. Conversely, extra sequence can arise through non-templated nucleotide addition, RNA ligation, or template jumping [100,101]. RtcB-mediated ligation, for example, generates U6–L1 chimeric RNAs that can account for chimeric insertions [101,116]. Finally, self-cleaving ribozymes can define template termini, as predicted for the 5′ ends of R2 transcripts [117,118], and observed more broadly across diverse retrotransposon systems [119], making RNA-structure-aware analysis important for accurately predicting template RNA termini.

For extant genomic insertions, care must first be taken to align each insertion to the appropriate family- or subfamily-specific reference, because closely related retrotransposons can share substantial sequence homology. Divergent lineages of the same family can coexist within a genome [120], and use of an incorrect reference can obscure insertion boundaries or lead to misannotation, as demonstrated for primate L1 subfamilies [121,122,123]. Even with accurate family assignment, however, identifying the genomic source locus for an individual insertion is often difficult because non-LTR retrotransposons are typically multicopy. Candidate retrotransposition-competent loci can be identified from genome sequence records by full-length retrotransposon sequence, intact ORFs with preserved catalytic amino acid motifs, and appropriate regulatory sequences, and then tested experimentally [124,125]. More direct source assignment can come from sequence carried into the new insertion: uniquely mappable 3′ or 5′ transduced sequence can identify the source locus [108,109,126], while source-specific single-nucleotide variants or haplotypes within the element can match offspring insertions to candidate genomic copies [63,127].

Trans-mobilized RNAs pose an additional problem because the resulting insertion does not identify the autonomous element that supplied the retrotransposition machinery. If RNP formation is governed by retrotransposon protein recognition of known structural or sequence features at the RNA 3′ end, common sequence features can identify candidate autonomous/nonautonomous element pairs [128,129,130,131]. For non-LTR retrotransposons with relaxed, poly(A)-directed RNP formation [132], autonomous/nonautonomous element pairs can be suggested by similar insertion junction signatures [133,134]. Experimental validation is required to definitively establish these autonomous/nonautonomous pairs [81,82,83,135].

5.3. Insertion Reaction: Target-Site Selection and First-Strand Cleavage

Below, we consider five determinants of target site selection (Figure 2B): (1) retrotransposon protein recognition specificity, (2) template-RNA-guided specificity, (3) host factor interactions, (4) target-site architecture, and (5) regional genomic context. Here, target-site architecture refers to the local physical structure of the DNA at the insertion site, including nicks, overhangs, and replication-associated structures such as Okazaki fragments. Regional genomic context encompasses broader features of the surrounding locus, including chromatin state, transcriptional activity, and replication timing. Importantly, these five determinants of target-site selection are not fully predicted by simple sequence-motif-based identification of target sites. More broadly, non-LTR retrotransposons fall along a spectrum of insertion-site specificity from highly site-specific to broadly dispersed rather than binary categories of site-specific vs. site-non-specific [70,136].

Biochemical mapping of EN–DNA specificity for R1Bm and human L1, together with concordant genomic insertion-site preferences, established endonuclease sequence specificity as the primary model for non-LTR retrotransposon target-site selection [21,133,137,138]. More direct evidence came from EN-domain swapping experiments between SART1 and TRAS1, which transferred target specificity between the two retrotransposon proteins [139]. Additionally, EN-independent L1 insertions lacked the canonical L1 target motif, further supporting a central role for EN sequence specificity [91]. For R2p, off-target mapping implicates nucleotide sequence outside the EN-contacted region as determining specificity [22,23,38], although not fully accounting for its unusually high degree of specificity, which indicates that the biochemical principles of target-site selection remain incompletely resolved.

Beyond protein determinants of specificity, base-pairing between the template RNA and the DNA 3′ end generated by first-strand cleavage can impose additional sequence requirements on TPRT initiation. For R1Bm, the addition of downstream 28S rRNA sequence to the RNA template increased retrotransposition efficiency and 3′ junction precision, consistent with a role for RNA–DNA complementarity [140]. More directly, mutation of the R2 target-site sequence disrupted in vitro TPRT from a prenicked DNA substrate, but activity was restored by compensatory changes to the 3′ tail of the template RNA [23]. Whether RNA sequence or structure contributes to target-site specificity beyond these base-pairing requirements has not been demonstrated.

A clear demonstration of host factor interaction as a determinant of specificity for non-LTR retrotransposons is Dictyostelium discoideum TRE5-A interaction with RNA polymerase III (Pol III) promoter-associated transcription complexes. Despite a high degree of site specificity for Pol III promoters, the exact position of insertion within the promoter sequence varies, suggesting recognition of transcription machinery rather than a specific DNA sequence, as validated through perturbative assays [141,142]. TRE5-A targeting through host factor interaction is analogous to previously characterized recruitment of some Ty LTR retrotransposons to Pol III-transcribed genes by interaction with transcription machinery [143]. As another potential mechanism of host factor influence, the interaction of L1ORF2p with PCNA could bias target-site selection toward lagging-strand templates through repeated PCNA loading during Okazaki fragment synthesis [144,145,146]. However, whether PCNA contributes to L1 target-site selection through this mechanism has not been tested directly.

Local DNA architecture has also been proposed as a determinant of target-site selection, though it has been demonstrated only in the context of L1ORF2p-mediated de novo insertions [45,145]. Mapping of de novo EN-dependent L1 insertions revealed an apparent bias for first-strand nicking in lagging-strand templates at the replication fork, consistent with biochemical preferences for cleavage substrates [35,45,145,147]. In parallel, mapping of de novo EN-independent L1 insertions revealed a bias for the opposite strandedness at replication forks, consistent with priming from Okazaki fragment 3′ ends [45].

The determinants of target-site specificity reviewed so far—those mediated by retrotransposon proteins, retrotransposon RNA, retrotransposon protein interaction with host factors, or target-site DNA architecture—describe predominantly local effects. Regional genomic context can also influence targeting. For human L1, de novo insertion mapping identified no consistent bias for insertion in transcribed regions or open chromatin, in sharp contrast to integration by retroviruses and DNA transposases [35,45]. Also, in contrast to human L1 but more in accord with other mobile genetic elements, de novo D. melanogaster I factor insertions showed a preference for targeting open, transcriptionally active chromatin [148,149].

While fixed genomic insertions provide some information about target-site selection, de novo insertion assays enable more direct and comprehensive identification of target-site preferences. Signatures of target-site sequence preferences can be preserved at fixed genomic insertions [133,150,151], but the contributions of broader genomic context can be more difficult to infer retrospectively. Purifying selection and rapid turnover at repetitive target loci can alter the observed genomic distribution of surviving insertions [35,64,67,149,152]. Host-mediated repression of inserted retrotransposons can also alter local chromatin, complicating inference of the chromatin state at the time of insertion [153]. De novo insertion mapping therefore enables more direct assessment of targeting preferences, both of local target-site sequence and of broader genomic context.

Genome-wide specificity claims require unbiased sampling, explicit detection-sensitivity estimates, retention and reporting of low-frequency events, and quantification of off-target insertions relative to a comprehensive total of all independent insertions. PCR of candidate off-target-site loci alone cannot establish genome-wide specificity. Recent studies adapting non-LTR retrotransposon machinery into gene insertion tools have not uniformly met these standards, in their reliance on candidate-site PCR, exclusion of low-frequency events by bioinformatic filtering, or claims of high specificity without explicit quantification [43,44,89]. See Table 1 for an overview of sequencing methods.

5.4. Insertion Reaction: DNA Primer Generation and Initiation of Reverse Transcription

Generation of the DNA primer for cDNA synthesis can occur by EN-mediated first-strand nicking, or a primer can be co-opted from a preexisting 3′ DNA end (Figure 2C). Early models of the non-LTR retrotransposon insertion mechanism often posited that preexisting DNA breaks could provide 3′ hydroxyl ends to prime reverse transcription [11,154]. Though biochemical studies later demonstrated that EN domain activity is sufficient to generate 3′ hydroxyl ends for TPRT [5,21,137], EN-independent priming from spontaneous DNA breaks and telomere ends is now also established as physiological [42,91,155]. EN-independent priming from Okazaki fragment 3′ ends has also been proposed [45]. The absence of a canonical target-site motif, large-scale target-site rearrangements, and/or internal initiation of reverse transcription can identify candidate EN-independent insertions [45,91]. Because EN-independent insertions exhibit distinct target-site signatures, candidate EN-independent events should be separated from canonical insertions when inferring EN-mediated target-site specificity. Although less deeply characterized than EN-deficient L1ORF2p, EN mutants of avian R2p and R1Bm exhibit diminished but measurable retrotransposition activity [23,140]. Bioinformatic analysis of these insertions suggests altered target-site preferences consistent with RT priming from preexisting 3′ ends [140].

The generation or capture of a DNA primer is only the first requirement for TPRT: productive reverse transcription additionally requires positioning of the RNA template relative to the primer 3′ end. EN-dependent TPRT also requires transfer of the newly generated 3′ end from the EN active site to the RT active site. For some non-LTR retrotransposon proteins, base-pairing of the RNA template to the DNA primer end (terminal complementarity, Figure 2C) is required for initiation [23]. Statistically enriched microhomology at 3′ junctions of extant insertions has suggested base-pairing requirements for RT initiation by taxonomically diverse non-LTR retrotransposon proteins [95,129,131]. In de novo insertion assays, perturbation of template RNA 3′ ends has demonstrated requirements of terminal complementarity for both precision and efficiency of TPRT initiation by R1Bm, SART1Bm, R2Ol, avian R2p, and R7Ag [23,140,156,157,158]. Terminal complementarity requirements are not universal, however: rare human L1 insertions are consistent with internal priming within the RNA template (Figure 2C) rather than initiation at its 3′ terminus [126,159], illustrating flexibility in primer–template positioning. Also, if priming is challenged by inadequate complementarity or insufficient length of single-stranded primer for stable TPRT initiation, extra nucleotides can be added prior to template copying (Figure 2C) [112,160,161].

5.5. Insertion Reaction: First-Strand cDNA Synthesis

Relative to the reaction steps considered thus far, bioinformatically reconstructing first-strand cDNA is particularly challenging. As discussed above (Section 5.2), many different processes before, during, and after first-strand synthesis can alter the final inserted sequence observed. Our analysis will describe historical efforts to disentangle these processes and indicate where future bioinformatic inquiry could provide important contributions.

A central unresolved question is what generates 5′ truncations, which could arise from RNA template degradation, premature termination of reverse transcription, nuclease cleavage of the cDNA, cDNA loss during 5′ junction formation, or a combination of these processes (Figure 2D–F) [162]. Though the hypothesis that 5′ truncation derives from premature termination by an insufficiently processive RT is pervasive in the literature (Figure 2D) [162], biochemical studies of R2 and L1 RTs do not provide support for this conclusion [102,163]. Additional investigation of this hypothesis will be informative for optimization of the genome-engineering applications for non-LTR retrotransposon proteins. Inherently, RNA is prone to hydrolysis and, in cells, to partial degradation (Figure 2D). RNA structural elements or base substitutions designed to improve the biological stability of a template yield notable suppression of 5′ truncation in R2p-mediated insertions, influencing both the percentage of full-length insertions and the amount of sequence lost as a truncation [23,105]. Finally, the cDNA product can be truncated during 5′ junction formation when second-strand synthesis is initiated on an internal region of first-strand cDNA (see Section 5.6 below) [61,62,162,164].

Irregularities in cDNA sequence architecture—including 5′ inversions, internal deletions, and internal duplications—are all candidates for first-strand synthesis-connected processes. Inversions at the 5′ end have been observed in ~20–25% of recently fixed L1 insertions, motivating the twin priming model (Figure 2D), in which reverse transcription primes from both ends of a DNA break at different positions on the same RNA template [165]. Inversion-containing de novo L1 insertions frequently exhibit architectures consistent with twin priming, although the proposed mechanism has not been directly demonstrated [48,49,63]. Similar inversions have not been reported for insertions by other non-LTR retrotransposons, raising the possibility that twin priming may be specific to L1ORF2p because of its tolerance for limited primer–template complementarity [95,159,165]. The occurrence of 5′ inversions in L1ORF2p-mediated de novo insertions varies by host, implicating host-dependent steps in their generation or resolution [61]. Other cDNA irregularities are less mechanistically resolved. Internal deletions and duplications have been observed within non-LTR retrotransposon insertion products [130,166], including de novo insertions [23,48], and therefore could arise from aberrant first- or second-strand synthesis rather than a post-insertion DNA rearrangement.

A final class of noncanonical first-strand synthesis events involves reverse transcription of host RNAs following reverse transcription of the intended template RNA [23,48,116]. While these events are consistent with end-to-end template jumping during first-strand synthesis (Figure 2D) [100], they may also reflect reverse transcription of a pre-ligated chimeric RNA [101].

An informative avenue for future studies would be to make systematic measurements of first-strand cDNA synthesis error rates, which remain poorly characterized for non-LTR retrotransposon RTs in cellular context. Sequencing of de novo L1ORF2p- and avian R2p-mediated insertions has yielded substitution rates on the order of 10⁻⁴ per base [23,48], similar to in vitro estimates for L1ORF2p and R2p [167,168]. However, insertion endpoint sequencing convolves errors from template RNA synthesis, first- and second-strand DNA synthesis, and sequencing, precluding direct measurement of first-strand synthesis fidelity in approaches applied to date. In vitro measurements are not a substitute, because they depend on assay conditions including dNTP concentrations and template context. Accurate measurement would require analysis of first-strand cDNA intermediates before second-strand synthesis and should employ low-error consensus sequencing.

5.6. Insertion Reaction: Second-Strand Cleavage, 5′ Junction Formation, and Second-Strand Synthesis

The mechanistic details of insertion completion—second-strand cleavage, 5′ junction formation, and second-strand cDNA synthesis—remain largely unresolved. Sequence-based inference is particularly challenging because endpoint features such as target-site alterations and 5′ truncations may report on multiple steps in the completion process rather than a single biochemical event (Figure 2E,F). De novo insertion studies provide the greatest opportunity to disentangle these processes through experimental perturbation, but new insertion 5′ junctions have often been sequenced far less deeply than 3′ junctions [22,35,38,45]. Mechanistic analysis of completion benefits from the characterization of both insertion junctions, ideally while retaining information about which junctions derive from the same insertion event.

Non-LTR retrotransposon insertions are commonly associated with TSDs, consistent with staggered nicks by the EN domain that generate 3′ overhangs [169,170]. Comparative analyses across host species show that TSD length patterns are more strongly associated with retrotransposon identity than with the host, implicating EN activity as a major determinant of TSD length [61,95]. However, host context can substantially alter the distribution of target-site outcomes even when the modal TSD length is preserved [61,93,97]. As an additional complication, TSDs generated by de novo L1 insertions generally match those observed in fossilized genomic insertions, but there is a much broader distribution of outcomes in the former, for reasons that have not yet been identified [48].

In contrast to most non-LTR retrotransposons, R2 insertions commonly have a 2-bp target-site truncation (TST) instead of a TSD [93]. Early biochemical reconstitutions of nicking by R2Bm protein showed that first- and second-strand cleavage positions matched the predominant junctions of extant B. mori insertions [5], with a 2-bp stagger generating 5′ rather than 3′ overhangs (Figure 2E). More recent studies of avian R2p de novo insertions directly linked R2p-mediated second-strand nicking to 5′-junction architecture and showed that precise cleavage and 5′ junction formation requires the recruitment of a second R2p subunit [46]. R2p variants deficient for second-strand cleavage could still produce full-length gene insertions, but with much more extensive target-site alterations, commonly ~1-kb TSDs [46,171]. Because R2 is the only system in which the retrotransposon EN has been shown to nick both DNA strands in cells, whether typical TSD formation is directly specified by retrotransposon-encoded cleavage activity has not been demonstrated. As an additional complication, both extant and de novo R2p-mediated insertions are often associated with TSTs that are slightly longer than the predicted 2-bp TST [62,97]. This likely occurs through nucleolytic removal of the 3′ overhang generated by 5′-to-3′ strand resection from the target site by host factors (Figure 2E) [62].

Heterogeneity in target-site alterations is accompanied by substantial variation in insertion architecture, particularly at the 5′ end. Full-length and 5′-truncated insertions are associated with different target-site alterations, consistent with distinct completion mechanisms [48,54,55,62]. Across vertebrate hosts, insertions by L1, L2, and Nimb proteins differ in patterns of 5′ truncations and in the use of microhomology for junction formation [28,29,61,95]. Distinct target-site alterations are associated with characteristic 5′-junction signatures, including microhomology and extra nucleotides (Figure 2F), suggesting that different DNA-end structures generated by variable second-strand cleavage may channel insertion intermediates into different completion pathways [28,55,62]. The absence of microhomology at most full-length L1 insertion 5′ junctions led to a model in which the RT jumps from the L1 RNA template to upstream target DNA (Figure 2F) [55], analogous to template jumping characterized for R2 RT [100,102].

Direct perturbation of host DNA repair has provided stronger evidence for multiple alternative completion pathways, although the mechanisms resolved so far may be host-, cell-type-, or element-specific. De novo ZfL2-2 insertions in canonical non-homologous end-joining-deficient DT40 cells were less frequently 5′-truncated [164]. The most detailed mechanistic picture has come from R2p-mediated insertion in human cells, where three repair pathways were associated with distinct 5′-junction outcomes: 53BP1–Shieldin–CST–Polα-primase fill-in with severe 5′ truncation, DNA polymerase theta-mediated end-joining with microhomology use and less severe truncations, and resection-dependent strand annealing with full-length insertions and precise junctions [62]. Unlike ZfL2-2 in DT40 cells, R2p insertion showed no dependency on canonical non-homologous end-joining factors [62], indicating that completion mechanisms can differ across element–host combinations. These findings motivate systematic reevaluation of 5′-junction architectures across extant and de novo non-LTR retrotransposon insertions using updated DNA repair models.

The final step for completing a non-LTR retrotransposon insertion is second-strand synthesis, which is largely invisible in analysis of endpoint insertion outcomes. Consequently, no complete molecular picture of second-strand synthesis has been drawn for any non-LTR retrotransposon insertion system, though involvement of host polymerases has been demonstrated for R2p [62]. Isolated claims of in vitro second-strand synthesis by non-LTR retrotransposon proteins exist [172,173], but relevance of this proposed mechanism to insertion in cells has not been demonstrated. Resolving this final step will likely require approaches that interrogate synthesis directly, such as analysis of reaction intermediates, measurement of host factor recruitment, or detection of polymerase-specific mutational signatures.

5.7. Products: Large-Scale Insertion-Associated Genome Rearrangements

Beyond the local target-site alterations typically associated with non-LTR retrotransposon insertion, larger-scale genome rearrangements can also result, including deletions, duplications, translocations, and inversions. Early studies identified recent D. melanogaster I factor insertions at deletion and inversion breakpoints, suggesting an association between I factor activity and genome rearrangements [24]. However, it could not be determined whether these rearrangements were coincident with insertion or resulted from subsequent ectopic recombination. Later de novo L1 assays recovered frequent deletions, duplications, and inversions at insertion sites [47,48,49], although drug selection over weeks of cell culture likewise prevented definitive assignment of these rearrangements to the insertion event itself.

Comparative genomic analysis subsequently identified rare surviving LINE- and SINE-associated rearrangements ranging from kilobases to candidate megabase-scale events [166,174,175,176], although purifying selection likely removes many such alleles from populations [64]. More direct evidence has come from cancer genomes and de novo insertions. L1 expression and retrotransposition are frequently elevated in cancers, in part through the loss of epigenetic and transcriptional repression [177,178]. Pan-cancer whole-genome sequencing revealed megabase-scale deletions and duplications as well as translocations and inversions associated with L1 retrotransposition [76]. Long-read reconstruction further suggested that some L1-associated translocations and inversions arise through the joining of concurrent retrotransposition intermediates [63], while rearrangement-specific reporter assays demonstrated that such events can occur coincident with L1 insertion [179]. Together, these studies establish L1 retrotransposition as a substantial source of large-scale genome rearrangements in tumors.

Whether comparable rearrangements are common across other non-LTR retrotransposons remains poorly understood. De novo insertion assays and analysis of recent insertions in diverse species will be needed to address this question. For elements such as R2 that target tandemly repeated, difficult-to-assemble loci, detecting rearrangements may require long-read sequencing supplemented by copy-number analysis or cytological methods. I factor-associated rearrangements in D. melanogaster and copy-number loss of PRINTed rDNA arrays in human cells nevertheless suggest that genome rearrangements are not unique to L1 [23,180].

5.8. Products: Abortive Reactions and Collateral Genome Damage

Abortive insertion reactions can occur without leaving a cDNA trace. Such events are therefore invisible to detection by molecular archaeology. For de novo insertion assays as well, standard insertion-mapping approaches generally do not capture collateral genome damage caused by EN activity that does not yield a recognizable TPRT product. Recent work has begun to address this blind spot using reporter-based detection of EN-dependent deletions lacking L1 sequence and genome-wide sequencing of L1-associated rearrangements, including junctions without an inserted L1 product [181,182].

Common methods for detecting or quantifying general genome damage often rely on γH2AX as a marker of DSB formation [183]. Although damage markers are not a direct or comprehensive indicator of DSBs, they provide a sensitive and scalable readout of genome-wide damage without requiring prior knowledge of break sites or the recovery and sequencing of products [184,185]. As non-LTR retrotransposon-based genome supplementation technologies approach clinical development, analysis of collateral damage, including damage not marked by a cDNA insertion, is increasingly of interest. Precedent from CRISPR–Cas9 genome editing illustrates how standard genotyping can substantially underestimate DNA damage: large deletions missed by short-range assays were recognized years after widespread adoption of a supposedly precise method, followed by evidence of chromosome-scale rearrangements and chromothripsis [186,187]. In assays of highly site-specific retrotransposon machinery, with R2 as representative, genomes with recent de novo insertions could be deep-sequenced for target loci lacking insertions as well as for general genome damage. This would reveal any repair signature at double-nicked target sites other than the expected re-ligation of the two-nucleotide overhangs, analogous to detection of DSB repair outcomes at Cas9 target loci [188]. For retrotransposon proteins that make non-site-specific insertions, single-cell or clonal cell line whole-genome sequencing may be necessary to detect global signatures of genome damage; this approach would also allow detection of collateral damage from site-specific insertion machinery.

6. Conclusions

In this review, we outlined historical examples of and best practices for mechanistic inquiry through the analysis of non-LTR retrotransposon insertions and their genomic context. To best enable progress going forward, we propose a standardized set of analyses beginning with a careful reconstruction of the molecular inputs needed to interpret insertion architecture and extending to a quantitative analysis of both junctions and internal sequence features. Reconstruction of the pre-insertion target site, mobilized template, junctions, internal architecture, and associated rearrangements can provide mechanistic information about successive steps of the insertion reaction. Applying this framework across different elements, hosts, and experimental perturbations should help distinguish broadly conserved features of TPRT from element- or host-specific mechanisms, while also improving evaluation of emerging retrotransposon-based genome-engineering systems. A more comprehensive characterization of insertion architecture, beyond identification of the element, target-site motif, and target-site duplications, can substantially expand the mechanistic information obtained from insertion datasets.

Interpretation of insertions as molecular records nevertheless requires caution because endpoint architectures are not necessarily unique to a particular mechanism. Rather than reporting representative examples or modal behavior, studies should report quantitative distributions of outcomes, including complex and difficult-to-resolve events. Conclusions must also account for biases introduced by experimental design, sequencing, and bioinformatic analysis. Robust mechanistic inference often requires recurrent patterns across independent events, experimental perturbation, or orthogonal evidence, and strong causal claims benefit from combining these approaches.

Recent advances provide substantial opportunities to extend mechanism-guided analysis beyond the small number of non-LTR retrotransposons that currently dominate mechanistic understanding. Higher-quality genome assemblies and long-read sequencing increasingly permit reconstruction of complete insertion events and repetitive target loci, while improved de novo insertion systems enable direct perturbation of retrotransposon components and host pathways. Together, these approaches should enable comparative analysis across a broader diversity of element architectures and host backgrounds and help resolve poorly understood aspects of retrotransposition, particularly the mechanisms of insertion completion. More standardized analysis and reporting will, in turn, make these comparisons increasingly informative across studies. As the diversity and scale of insertion datasets continue to grow, treating each insertion as a molecular record of the reaction that generated it should allow sequence analysis to play an increasingly direct role in defining the mechanisms of non-LTR retrotransposition.