Long-Read RNA Sequencing Reveals Active Transposable Elements in Human Cell Types

A research team led by Chaemin Lim, Hyunsu An, and Jihwan Park has established a comprehensive reference annotation of locus-specific, transcriptionally active transposable element transcripts across the human genome by analyzing 2,596 publicly available long-read RNA-sequencing datasets. Published in Genomics & Informatics, the study demonstrates that approximately one out of every nine transposable element sequences cataloged in standard repeat databases is actively transcribed as a putatively functional transcript within human cells. Transposable elements have historically been categorized as nonfunctional genomic remnants, but technological constraints have long obscured their expression at unique genomic positions across distinct tissues. By harnessing long-read data spanning 21 organs and 71 cell lines from healthy donors and diseased patients, the authors assembled full-length transcripts to systematically characterize locus-specific transposable element activity. Their resulting framework reveals that transposable element expression is widespread, structurally complex, and tightly regulated across human cell types, providing an empirical tool to investigate repeat-derived RNAs in physiological development and human disease.

Transposable elements constitute nearly 50 percent of the human genome and were once dismissed as junk DNA, but sequencing studies have revealed their critical contributions to gene regulation, chromatin architecture, and pathological states. During mammalian development, specific transposable element families display stage-specific expression profiles and interact with transcription factors to coordinate key developmental genes. In particular, long terminal repeat retrotransposon transcripts participate in epigenetic remodeling by shaping the three-dimensional structural organization of chromatin. Conversely, dysregulated transposable element activation can trigger deleterious consequences, such as onco-exaptation, in which a transposable element promoter aberrantly drives oncogene expression during tumorigenesis. While single-cell RNA sequencing has accelerated cellular investigation, existing computational tools like scTE group elements at the family level without genomic coordinates, whereas methods like SoloTE rely on mutation predictions rather than observed transcription. Furthermore, long-read resources like CELLO-seq remain restricted to mouse blastomeres, leaving human biology without a comprehensive annotation of experimentally confirmed, locus-specific transposable element transcripts.

A primary motivation for developing an empirical long-read annotation stems from the structural shortcomings of traditional repeat identification tools. Standard annotations from the UCSC Genome Browser RepeatMasker database provide a valuable catalog of repeat loci, yet RepeatMasker was originally engineered to mask repetitive segments to facilitate short-read alignment and conventional non-repeat gene analysis. Consequently, RepeatMasker lacks transcript-level architecture, coordinate exon models, and direct transcriptional activity data, meaning its entries frequently include ancient, fossilized elements that are completely silent in living cells. In short-read and single-cell RNA sequencing, mapping ambiguities frequently cause false-positive alignments to these inert repeat sequences. Long-read sequencing platforms, by contrast, generate continuous sequence reads extending from several kilobases up to hundreds of kilobases, enabling direct sequencing of intact full-length RNA molecules. This capability allows researchers to span complex repetitive elements and resolve ambiguous genomic alignments that have historically prevented accurate repeat quantification across human cell types.

To build the reference, the authors collected 2,596 sequencing experiment runs from the NCBI Sequence Read Archive using targeted queries including human, long-read, rna-seq, nanopore, and pacbio. The team retrieved raw FASTQ files using SRA Toolkits version 3.0.0 and aligned them to human reference genome GRCh38 using minimap2 version 2.24 with the splice preset and default parameters. Alignments were processed with Pysam version 0.19.0, applying a mapping quality threshold of 30 and excluding hard-clipped reads because minimap2 realigns poorly mapped sequences using soft clipping while truncating alignments through hard clipping. To isolate authentic transposable element activity, the team constructed whole-genome binary masks with the bitarray version 2.5.1 package, assigning a value of one to coordinates matching RepeatMasker elements and zero elsewhere, alongside a parallel mask representing reference gene exons. Long-read sequences overlapping non-TE repetitive structures, such as transfer RNAs and simple repeats, were removed, ensuring only verified transposable element reads advanced into downstream transcriptome reconstruction.

The researchers categorized transposable element-overlapped reads into four distinct groups based on genomic positions relative to known non-TE genes and repeat boundaries. The first category encompassed intergenic transcripts overlapping solely with intergenic transposable element regions without non-TE gene overlap. The second category comprised intragenic transcripts overlapping non-TE genes while exhibiting transposable element coverage exceeding 80 percent of aligned read length. The third category captured fusion transcripts overlapping both annotated non-TE genes and adjacent intergenic transposable element sequences, representing potential chimeric transposable element-gene fusions. A fourth catch-all category gathered remaining reads that ambiguously spanned both transposable element and non-TE gene regions. Across the 2,596 datasets, transposable element reads accounted for 15.6 percent of uniquely mapped reads, with the ambiguous fourth category comprising 12.3 percent. To prevent false-positive transcript models caused by uncertain genomic origins, the authors excluded this fourth category and assembled remaining reads into transcripts using Stringtie2 version 2.1.7 with parameters -L -G -v -p 12 -j 1 -g 0 -f 0.01 -l STRG -m 50 -t -c 1.5 -s 1.5, guided by GENCODE release 105 and RepeatMasker.

The assembled transcriptome yielded 535,724 locus-specific transposable element-derived transcripts, establishing that one out of every nine elements among the 4,891,349 transposable elements in RepeatMasker is transcriptionally active. Approximately 94 percent of these transcripts, representing 504,100 individual models, possessed only a single isoform, whereas multi-isoform transcripts were primarily generated by protein-coding superfamilies including LINE, ERV, DNA transposons, and chimeric gene fusions. Evaluating superfamily distributions revealed substantial differences between RepeatMasker baselines and actual transcription: LINE elements and long terminal repeat retrotransposons accounted for 44 percent and 19 percent of the new catalog compared to 32 percent and 15 percent in RepeatMasker, while short interspersed nuclear elements, DNA transposons, and retroposons exhibited reduced representation at 29 percent, 7 percent, and 0.3 percent versus 39 percent, 10 percent, and 0.1 percent. Validating transcript splicing with Logomaker confirmed that canonical AG acceptor and GT donor motifs were highly conserved across transposable element transcripts, even when analyzing models entirely independent of non-TE genes.

Direct examination of assembled loci demonstrated the power of long-read sequencing to resolve unannotated transcript structures, alternative splicing isoforms, and revised gene boundaries. The authors identified novel autonomous transcripts, including STRG.501136, a 1,270-base-pair multi-element transcript combining sequences from the AluSz, AluY, and FLAM-A subfamilies of the SINE superfamily. In another locus, the ERV transcript STRG.437231 generated three distinct full-length splicing isoforms aligning with intact proviral sequences in ERVmap, whereas RepeatMasker had fragmented this single active locus into nine separate annotations encompassing two LTR7, six HERVH, and one MER4B elements. Furthermore, the annotation refined transcription end sites, demonstrating that an isoform of the oncogenic long noncoding RNA LINC-ROR terminated at position 57,062,614 base pairs on chromosome 18 rather than at 57,063,555 base pairs as indicated in GENCODE, with long-read sequencing reads directly confirming the revised site. In addition, the catalog recovered known oncogenic chimeric fusions, identifying the LOR1a-IRF5 fusion transcript STRG.482309, which is enriched in B-lymphoblastoid cell lines and linked to abnormal lymphocyte proliferation.

To evaluate single-cell performance, the investigators applied their transposable element reference to single-nucleus RNA-sequencing data from the Genotype-Tissue Expression project, examining eight human tissues—breast, esophagus mucosa, esophagus muscularis, heart, lung, prostate, skeletal muscle, and skin—obtained from 16 healthy donors. The team downloaded 102 barcoded BAM files totaling 955 gigabytes using gsutil version 5.5 and processed cell-level alignments with Pysam and interval tree indexing version 3.1.0, filtering out transposable element transcripts overlapping GENCODE non-TE genes as a conservative measure. Downstream analysis of 174,419 cells using Scanpy version 1.9.1 revealed that transposable element-derived transcripts constituted an average of 6.7 percent of the total cellular transcriptome across tissues. Skeletal muscle and lung exhibited the highest transposable element proportions, exceeding 10 percent of expressed transcripts, while 12 transposable element transcripts were detected across all eight tissue types. LINE elements dominated the transposable element transcriptome in all tissues with an average share of 58.5 percent, and sequencing depth comparisons confirmed that elevated transposable element detection in muscle and lung was not driven by unique molecular identifier counts.

At single-cell resolution, the GTEx dataset separated into 74 distinct cell clusters corresponding to established cell annotations when profiled using combined non-TE genes and transposable element-derived transcripts. Differential expression analysis performed with the Wilcoxon rank-sum test at an adjusted p-value threshold below 0.05 identified 16,459 differentially expressed transposable element-derived transcripts across these cell types. Skeletal muscle myocytes displayed the greatest number of differentially expressed transposable element transcripts, followed by alveolar type II epithelial cells and alveolar macrophages, with distinct cell-type-specific profiles also observed in cardiac myocytes, adipocytes, and T cells. When the 74 cell types were grouped into five major categories—myocyte, epithelial, endothelial, immune, and others—the myocyte and epithelial categories exhibited significantly higher mean transposable element transcript counts than the immune and other groups. These findings demonstrate that transposable element expression is uncoupled from simple sequencing depth artifacts at the cell-type level, highlighting its biological relevance as a signature of cellular identity.

Despite the high resolution achieved by large-scale long-read assembly, the investigators identified specific methodological limitations that warrant consideration in future genomic studies. In particular, default alignment parameters of minimap2 can cause under-detection of short micro-exons measuring fewer than 30 base pairs, meaning the conservative filtering approach utilized to maximize mapping specificity in repetitive regions may have reduced sensitivity for very small exon structures. The authors proposed that future iterations could incorporate specialized micro-exon aligners such as deSALT, optimize gap penalties, or integrate known splice junction annotations to further enhance transcript recovery. Nonetheless, by compiling 535,724 validated, locus-specific transposable element transcripts from 2,596 long-read sequencing datasets, the study establishes an expansive empirical resource that supersedes passive repeat masking. This comprehensive reference annotation equips the scientific community to analyze active transposable elements, detect novel isoforms, and investigate how repeat-derived transcripts regulate normal cellular identity, developmental programs, and human disease pathogenesis.

Subject of Research: Locus-specific annotation of transcriptionally active transposable elements using human long-read RNA sequencing

Article Title: Identification of cell-type-specific, transcriptionally active transposable elements using long-read RNA-sequencing data-based comprehensive annotation

Article References: Lim, C., An, H., & Park, J. (2025). Identification of cell-type-specific, transcriptionally active transposable elements using long-read RNA-sequencing data-based comprehensive annotation. Genomics & Informatics, 23(1), Article 17. https://doi.org/10.1186/s44342-025-00048-1

Image Credits: AI Generated

DOI: 10.1186/s44342-025-00048-1

Keywords: transposable elements, long-read RNA sequencing, transcriptome assembly, single-cell RNA-seq, gene expression, GTEx, alternative splicing, cellular identity, Identification, cell-type-specific, transcriptionally, active