Ancient Viral Fossils Decoded: New Tool untangles the Human Genome’s Endogenous Retroviruses

Buried within every human cell lies a molecular archive of ancient infections. Roughly eight percent of the human genome consists of human endogenous retroviruses, or HERVs, the fossilized remains of retroviruses that inserted themselves into the germline of our ancestors millions of years ago. Far from being inert genetic junk, these elements are now known to influence embryonic development, shape immune responses, and contribute to diseases ranging from multiple sclerosis to cancer. Yet studying them has long been one of genomics’ most stubborn technical challenges, because their repetitive sequences make it extraordinarily difficult to determine which of the thousands of scattered copies a given sequencing read actually came from. A new computational framework called ERVmancer, described in the journal Genome Biology by Andrew Patterson, Noam Auslander, and colleagues at The Wistar Institute, the University of Pennsylvania Perelman School of Medicine, and the NIH Neuroimmunology Branch, promises to change that by resolving the mapping ambiguity that has obscured HERV biology for decades.

The core problem is one of identity. Short-read sequencing technologies, which dominate modern transcriptomics, produce fragments of RNA sequence that are typically only 100 to 150 nucleotides long. When those fragments originate from a HERV locus, they often match dozens or hundreds of nearly identical copies scattered across the genome with equal fidelity. Standard alignment algorithms, faced with equally good candidates, either discard the read as ambiguous or assign it arbitrarily, which means that expression signals from individual proviral loci are blurred together or lost entirely. Long-read sequencing platforms can span entire proviruses and resolve their origins, but they remain more expensive, lower throughput, and less commonly available, which limits their use in large clinical cohorts. The result has been a persistent blind spot: researchers know that HERV transcripts are present in a sample, but they frequently cannot say which specific retroviral elements are active, where they sit in the genome, or what regulatory logic controls them.

ERVmancer attacks this ambiguity by borrowing an idea from evolutionary biology. Rather than treating each HERV copy as an isolated entity, the framework organizes the 3,043 HERV loci it tracks into a phylogenetic structure, grouping related proviruses into clades that reflect their shared ancestry. When a short sequencing read cannot be uniquely assigned to a single locus, the method uses the phylogeny to make a principled probabilistic assignment: reads that are ambiguous among closely related copies within a clade contribute to the quantification of that clade, while information that distinguishes individual loci is preserved wherever the sequence allows it. In this way, ERVmancer can report expression at multiple resolutions simultaneously, from the individual proviral locus up through families and entire clades, giving researchers a hierarchical view of retroviral activity that mirrors the evolutionary reality of these elements rather than flattening it.

To find out whether this phylogeny-guided strategy actually works better than existing approaches, the team benchmarked ERVmancer using matched datasets in which the same biological samples had been sequenced with both short-read and long-read technologies. The long-read data served as a ground truth of sorts, since it can resolve individual loci directly. Across these comparisons, ERVmancer outperformed existing methods in both sensitivity, meaning its ability to detect genuine HERV expression, and specificity, meaning its ability to avoid false assignments. The evaluation extended across HERV families, and supplementary analyses using simulated sequencing data further probed how reads were distributed among internal phylogenetic clades, providing a detailed picture of where the method succeeds and where uncertainty remains. The tool has been released as a Bioconda package, a distribution channel widely used in the bioinformatics community, which should make adoption straightforward for laboratories analyzing RNA sequencing data.

The practical value of the framework was demonstrated in two disease contexts with very different biology. The first was multiple sclerosis, an autoimmune disease of the central nervous system in which HERV activity has long been suspected of playing a role, though pinning down which elements are involved has been difficult. The researchers applied ERVmancer to RNA sequencing data from spontaneous lymphoblastoid cell lines derived from patients with multiple sclerosis and from healthy controls, all of whom provided written informed consent under protocols reviewed by institutional review boards at the National Institutes of Health and The Wistar Institute. The analysis reproduced expression patterns that had previously been observed only with long-read sequencing, validating that the short-read approach could recover biologically meaningful signal, and it identified HERV expression patterns specific to the disease context, including variability within the ERV3 clade that the team confirmed independently using quantitative reverse-transcription PCR.

The second application turned to cancer, where the interplay between tumor suppressors and endogenous retroviruses has become an increasingly active frontier. Analyzing breast cancer cell lines from the Cancer Cell Line Encyclopedia, ERVmancer uncovered a striking regulatory relationship: the transcription factor p53, famous as the guardian of the genome for its role in suppressing tumors, appears to actively suppress the expression of the HERVH-LTR7 clade. This finding is significant because HERVH and its LTR7 long terminal repeat are known to be active in embryonic stem cells and in certain tumors, where they help maintain a stem-like, proliferative state. The observation that p53 exerts suppressive control over this clade suggests that one way the cell’s primary tumor suppressor keeps cells from reverting to a primitive, uncontrolled state is by silencing the viral remnants that normally fuel such programs in early development. Genome-wide analysis of p53 binding sites across diverse HERV clades and families, presented in the supplementary data, supports the breadth of this regulatory reach.

These findings illustrate why the ability to resolve HERV expression at the clade and locus level matters. A bulk measurement that lumps all HERVH activity together might obscure the fact that a specific subset of proviruses, defined by their evolutionary lineage and their regulatory sequences, is being selectively repressed or activated. In multiple sclerosis, distinguishing disease-associated expression within particular clades could eventually clarify whether endogenous retroviral transcripts act as drivers of pathology, as biomarkers of immune dysregulation, or as byproducts of inflammation. In oncology, understanding which retroviral elements a tumor reactivates, and which tumor suppressors keep them in check, could inform strategies to exploit viral-like antigens on cancer cells for immunotherapy, an approach that has already attracted considerable interest because HERV expression in tumors can generate targets that distinguish malignant cells from healthy tissue.

The methodological advance also speaks to a broader trend in genomics: the recognition that repetitive regions of the genome, long excluded from analysis or collapsed into consensus sequences, contain biology worth recovering in full. As short-read sequencing remains the workhorse of large-scale studies, including biobank-scale transcriptomics and clinical cohorts, tools that can extract locus-resolved information from inherently ambiguous data multiply the value of existing datasets without requiring new sequencing campaigns. ERVmancer’s hierarchical output means that researchers with different questions can query the same analysis at different resolutions, whether they care about a single provirus with a known regulatory function or about the aggregate behavior of an entire retroviral lineage across hundreds of samples.

The study, which was published open access on 25 September 2026 and led by corresponding authors Andrew Patterson and Noam Auslander, brought together expertise spanning virology, neuroimmunology, and computational biology, including contributions from Steven Jacobson of the NIH Neuroimmunology Branch and Paul Lieberman and Samantha Soldan of The Wistar Institute. It was supported by funding from the W. W. Smith Charitable Trust, the Michelson Medical Research Foundation, the United States Department of Defense, and multiple institutes of the National Institutes of Health. As with any computational method, its conclusions will need to be tested and extended by independent groups, and the authors note that the published version followed full peer review at Genome Biology.

For a field that has spent decades peering at the ghostly traces of ancient viruses through a fog of repetitive sequence, ERVmancer offers something close to a lens correction. By letting the evolutionary relationships among HERVs guide the interpretation of ambiguous reads, the framework turns a longstanding liability of short-read data into a tractable inference problem, and it delivers early biological dividends: confirmation that long-read-quality insight can be recovered from ordinary transcriptomic data, disease-specific expression signatures in multiple sclerosis, and evidence that p53, the cell’s most celebrated tumor suppressor, doubles as a warden of the viral archive within our DNA. As researchers begin applying the tool to other cancers, autoimmune conditions, and developmental systems, the regulatory grammar of the genome’s viral fossils may finally come into focus, revealing how infections from the deep past continue to script human biology today.

Subject of Research: A phylogeny-guided computational method for quantifying human endogenous retrovirus expression and regulation in health and disease

Article Title: ERVmancer: a phylogeny-guided framework for decoding human endogenous retrovirus regulatory mechanisms in health and disease

Article References: Patterson, A., Duong, B., Yoon, L., Foster, M., MacMullen, L., Wickramasinghe, J., Lucas, A., Srivastava, A., Jacobson, S., Murphy, M. E., Soldan, S. S., Lieberman, P. M., & Auslander, N. (2026). ERVmancer: a phylogeny-guided framework for decoding human endogenous retrovirus regulatory mechanisms in health and disease. Genome Biology. https://doi.org/10.1186/s13059-026-04287-5

Image Credits: AI Generated

DOI: 10.1186/s13059-026-04287-5

Keywords: human endogenous retroviruses, HERVs, ERVmancer, phylogenetics, RNA sequencing, short-read mapping, multiple sclerosis, breast cancer, p53, HERVH-LTR7, transposable elements, computational genomics