Over the past three decades, numerous complementary approaches have been developed to identify and classify human endogenous retroviruses (HERVs). While these studies have greatly expanded our understanding of HERV diversity, they have also produced multiple partially overlapping annotation systems that differ in methodology, nomenclature, and evolutionary interpretation, making it difficult to compare equivalent loci across studies.

Here, we present HERVdb, a unified annotation resource built around a systematically reconciled evolutionary reference framework for human endogenous retroviruses. Rather than replacing existing annotation resources, HERVdb establishes a common evolutionary context through which annotations from diverse sources can be interpreted consistently while preserving explicit links to historical nomenclatures and supporting evidence. This framework is subsequently used to integrate complementary genome-wide annotations into a unified reference dataset of HERV loci linked to lineage classifications, structural features and legacy identifiers.

By establishing a unified reference dataset for human endogenous retroviruses, HERVdb provides a foundation for comparative and functional genomics, facilitating studies of ERV contributions to gene regulation, immunity, and disease. The framework established here is readily extensible to the annotation of endogenous retroviruses in other vertebrate genomes.

Construction of the evolutionary reference framework

To establish a unified evolutionary reference framework for HERVs, we compiled lineage definitions from all major published HERV classification schemes together with widely used genome annotation resources. Representative sequences, consensus elements, annotation library entries, and published cross-references were recovered from the literature and public databases. Rather than assuming similarly named entities represented equivalent evolutionary groups, each lineage concept was initially treated as an independent hypothesis. Relationships between historical classifications were resolved using published synonymies, shared reference sequences, comparative analyses, and maximum-likelihood phylogenetic reconstruction of representative reverse transcriptase (RT) sequences. For entities lacking recoverable RT regions, including LTR families and highly degraded proviruses, evolutionary relationships were inferred from published structural and phylogenetic evidence. The resulting framework reconciles multiple lines of evidence into a curated set of lineage-specific reference sequences representing the diversity of recognized HERV lineages, while preserving original aliases, representative sequences, and supporting evidence.

Genome-wide identification of HERV loci

Genome-wide HERV-derived sequences were independently recovered from the human genome using the DIGS platform with retroviral protein references derived from the curated HERV diversity defined by our evolutionary reference framework. Recovered coding regions were assigned to lineages by maximum-likelihood phylogenetic placement within the curated RT reference tree, supplemented by additional coding regions and lineage-specific reference information where RT sequences were absent. Individual gene features were consolidated into proviral loci using empirical genomic distance thresholds and subsequently refined using genomic organization, phylogenetic consistency, and LTR architecture. This process generated an independently reconstructed catalogue of 17,806 phylogenetically classified HERV loci.

Integration of existing annotation resources

The DIGS catalogue was integrated with RepBase and RetroTector to generate a unified genome-wide HERV resource. Each annotation system was first converted into a common locus-based representation, after which overlapping loci were reconciled using reciprocal genomic overlap while retaining the original annotations contributed by each resource. Because the three databases differ in detection strategies, locus boundaries, and lineage nomenclature, HERVdb preserves all original annotations rather than forcing agreement between systems. Instead, lineage assignments from all contributing resources were reconciled using the unified evolutionary framework, enabling consistent classification while preserving historical nomenclatures, alternative lineage assignments, and conflicting interpretations where they exist.

Hierarchical taxonomic framework

Of the 513,487 merged loci, 151,359 loci were assigned to a HERV or LTR lineage defined by our evolutionary reference framework and were subsequently incorporated into our standardized hierarchical taxonomy. HERVs are organized into Clades, Groups, Subgroups, Lineages, and Sublineages. Names follow the format:

Clade.Group.Subgroup.Lineage-Numeric ID

For example, the locus ERVE-4, associated to better immunotherapy responses in kidney cancer, is represented as ERV-1.Gamma.Eta.E-4, preserving legacy nomenclature while embedding each lineage within a consistent evolutionary hierarchy. Historical names are retained throughout HERVdb as searchable aliases.

Locus classification and curation

All loci in HERVdb were further classified according to structural composition and supporting evidence into High-Support HERVs, Low-Support HERVs, and solo LTRs. High-Support HERVs comprise loci supported by multiple annotation resources and containing substantial proviral structure, whereas Low-Support HERVs include fragmented or single-source annotations. The resulting High-Support set of 8,617 loci constitutes the primary HERV proviral reference dataset within HERVdb.