Tracing ancient viral footprints: a comprehensive study of endogenous viral elements in Bombus species

Abstract

Background

Endogenous Viral Elements (EVEs) are viral sequences integrated into the host germline and passed to offspring. Most virus types can integrate, often with the help of host retroelements, especially for non-retroviral RNA viruses. It is known that EVEs are widespread across insect species and related to an extensive range of virus taxa, many of which might share similar evolutionary origins. Bombus bees are essential pollinators that have been experiencing worldwide colony declines in recent decades. Therefore, uncovering genetic elements and pathways to better understand host–pathogen interactions is crucial in conserving biodiversity.

Results

Non-retroviral Integrated RNA Virus Sequences (NIRVS) were widespread in Bombus genomes, without a clear correlation between genome size and the number of EVEs. Most of the EVEs were single-copy, ranging from 111 to 3,729 bp with an average of 504 bp. Most of them share similarities with unclassified viruses and known viruses belonging to the families Partitiviridae and Virgaviridae, as well as the order Martellivirales. We observed that over 25% of the NIRVS contain conserved domains, with larger ones having a higher probability of functional annotation. Most NIRVS with conserved domains contained Polymerase-related motifs, the most represented group of domains among Bombus species. A comprehensive analysis of the NIRVS sharing pattern suggests that they are more likely to be inherited from a common ancestor than to result from integration events after speciation. Also, viral elements are widely conserved amongst species. Furthermore, we investigated transcriptional activity and the potential of the NIRVS to function as a priming agent for antiviral responses against exogenous viruses. On that note, most NIRVS in Bombus are transcriptionally active, and some share 15 nt of contiguity with exogenous bee viruses and could potentially be used as templates for piRNA production.

Conclusions

The integration of non-retroviral RNA viruses into bumblebee genomes is ancient and represents a dynamic evolutionary process in which many viral elements are conserved, shared, and may be functionally active in Bombus bees.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13100-026-00393-0.

Background

Endogenous Viral Elements (EVEs) are viral sequences integrated into host genomes and transmitted vertically to offspring due to insertion in the germline [1]. It is well known that viruses of all genome types may undergo insertion, driven mainly by the action of cellular retroelements, including non-retroviral integrated RNA virus sequences (NIRVS), which are often found in regions of repetitive DNA and are frequently flanked by Long Terminal Repeat (LTR) retroelements. The fact that reverse transcription for these elements occurs in the cytoplasm, is active in germline cells, and that some may harbor virus-like domains supports the association between non-retroviral RNA EVEs and retroelements [2].

Those insertions may evolve neutrally and accumulate mutations, losing their viral identity when inserted in silenced regions of host DNA, or be inserted in transcriptionally active areas where they could act as templates for the biogenesis of small RNAs (sRNAs) or even be translated into defective viral proteins, both of which can contribute to antiviral immunity. Some EVEs may even undergo exaptation, assuming different advantageous roles in their hosts [3]. Some examples of functional EVEs in animals include syncytin genes, elements derived from endogenous retroviruses that play a crucial role in placental development in mammals [4], and viral elements in microgastroid parasitoid wasps that are translated into bracovirus and ichnovirus virions to modulate host physiology and facilitate parasitism [5].

In mosquito genomes, the diversity and structure of EVEs have been screened [6–8], and a significant source of EVE transcription, related to the generation of sRNAs, has been confirmed [9]. Most importantly, a functional link between non-retroviral EVEs and antiviral immunity has also been tested and confirmed, in which antiviral piRNAs are produced in the presence of a naturally occurring EVE and its cognate virus, limiting its replication in mosquitoes [10].

This discovery provides further evidence of the potential crucial roles that non-retroviral EVEs may play in insect immunity. This potential offers new perspectives on how to mitigate infections and control natural viral hazards to key species in our survival, inspiring novel antiviral strategies, and also provides a better understanding of how these integrations have shaped eukaryote evolution. Beyond medically necessary species such as mosquito vectors to human arboviruses, we highlight in this study the need for a deeper understanding of the presence of EVEs in insect pollinators, specifically bees.

In addition, approximately 87,5% of all terrestrial angiosperms rely upon animals for pollination [11]. The total economic value of insect pollination worldwide is estimated at more than EUR 153 billion annually [12]. In the USA, bees contributed USD 15,12 billion [13] to the economic value of agricultural production. They are the primary pollinators of plants and visit more than 90% of the world’s leading 107 crop types [14].

Among managed and wild species, more than 20,000 bee species have been described worldwide [15]. The western honeybee (Apis mellifera), the eastern honeybee (Apis cerana), some bumblebees (Bombus spp.), stingless bees, and solitary bees are part of the small group of managed species (~ 50 species) across the world. Alongside them, there is evidence of the roles of wild pollinators and diverse pollinator assemblages in improving the quality and quantity of global crop production [16], which also provides crucial benefits for maintaining current biodiversity and ecosystem stability [17]. However, climate change, pollution, habitat loss, and pathogens, among others, are factors increasingly linked to colony loss worldwide, which could severely hinder human food security and disrupt ecosystem functions [18–21]. Thus, it is important to address the urgency of finding possible solutions to colony losses, mitigating these factors, in which viral pathogens are among the most concerning.

It is known that most bee viruses have single-stranded, positive-sense RNA genomes, including the globally significant Deformed Wing Virus (DWV), which replicates in many bee species, including honeybees and bumblebees. Kashimir Bee Virus (KBV), Israeli Acute Paralysis Virus (IAPV), and Acute Bee Paralysis Virus (ABPV) have also been identified in those bees and contribute to high mortality in Bombus populations [22–26].

Many bumblebees, Bombus spp. (Hymenoptera: Apidae), have experienced significant population declines over the last few decades, raising concerns about wild bee populations, agricultural production, and the maintenance of biodiversity as we know it [17, 27–39]. They are crucial pollinators, increasing pollination efficiency by vibrating their bodies [40], remaining active in cold weather, and being larger and hairy, which enables them to carry larger pollen loads [41].

This study aims to assess the presence and diversity of conserved non-retroviral RNA Endogenous Viral Elements in publicly available Bombus genomes to better understand the endogenous virome of bees and to enhance current knowledge of virus-host interactions, including potential antiviral activity from EVEs.

Methods

Retrieval of public data

Genomic data for forty-one bumble bee species were sourced from the National Center for Biotechnology Information (NCBI) GenBank database (https://www.ncbi.nlm.nih.gov/genome), with species details and metadata in Table 1. We also collected public Bombus spp. RNAseq libraries (Additional File 1). Furthermore, RefSeq viral proteomes were obtained from GenBank (< https://ftp.ncbi.nlm.nih.gov/refseq/release/viral/ > Accessed in March 2025).

Table 1.

Overview of Bombus genomes collected from the NCBI genome database

Organism Name	Taxon. ID	Assembly Accession	Release Date	Assembly Level
B. affinis	309941	GCF_024516045.1	05/08/2022	Chromosome
B. balteatus	85657	GCA_019201815.1	13/07/2021	Contig
B. bifarius	103933	GCF_011952205.1	03/04/2020	Scaffold
B. breviceps	395515	GCA_014825925.1	02/10/2020	Chromosome
B. campestris	207624	GCA_905333015.3	16/10/2022	Chromosome
B. confusus	217217	GCA_014737475.1	25/09/2020	Scaffold
B. consobrinus	130686	GCA_014737455.1	25/09/2020	Scaffold
B. cullumanus	2562068	GCA_014737535.1	25/09/2020	Scaffold
B. dahlbomii	85658	GCA_037178635.1	18/03/2024	Chromosome
B. difficillimus	395520	GCA_014737525.1	25/09/2020	Scaffold
B. fervidus	203811	GCA_041682495.1	03/09/2024	Chromosome
B. flavifrons	103934	GCF_040668555.1	12/07/2024	Scaffold
B. haemorrhoidalis	207636	GCA_014825975.1	02/10/2020	Chromosome
B. hortorum	85660	GCA_905332935.1	19/03/2021	Chromosome
B. huntii	85661	GCF_024542735.1	09/08/2022	Chromosome
B. hypnorum	30191	GCA_911387925.2	14/10/2022	Chromosome
B. ignitus	130704	GCA_014825875.1	02/10/2020	Chromosome
B. impatiens	132113	GCA_043295415.1	17/10/2024	Chromosome
B. jonellus	85663	GCA_964197665.1	03/07/2024	Chromosome
B. lapidarius	30192	GCA_964186655.1	07/06/2024	Chromosome
B. muscorum	203813	GCA_963971185.1	04/03/2024	Chromosome
B. opulentus	2024865	GCA_034509555.1	26/12/2023	Chromosome
B. pascuorum	65598	GCF_905332965.1	19/03/2021	Chromosome
B. picipes	309970	GCA_014737485.1	25/09/2020	Scaffold
B. polaris	130708	GCA_014737335.1	24/09/2020	Scaffold
B. pratorum	30194	GCA_930367275.1	13/02/2022	Chromosome
B. pyrosoma	396416	GCF_014825855.1	02/10/2020	Chromosome
B. rufofasciatus	309971	GCA_040285845.1	28/06/2024	Chromosome
B. sibiricus	421273	GCA_014737505.1	25/09/2020	Scaffold
B. skorikovi	395565	GCA_014737355.1	24/09/2020	Scaffold
B. sonorus	203818	GCA_029958995.1	09/05/2023	Scaffold
B. soroeensis	184059	GCA_014737365.1	24/09/2020	Scaffold
B. superbus	1869276	GCA_014737385.1	24/09/2020	Scaffold
B. sylvestris	30201	GCA_911622165.2	15/10/2022	Chromosome
B. sylvicola	309975	GCA_019677175.1	18/08/2021	Contig
B. terrestris	30195	GCF_910591885.1	09/02/2022	Chromosome
B. turneri	686820	GCA_014825825.1	02/10/2020	Chromosome
B. vancouverensis	2705178	GCF_011952275.1	03/04/2020	Scaffold
B. vestalis	30202	GCA_963556215.1	10/10/2023	Chromosome
B. vosnesenskii	207650	GCA_026744215.1	08/12/2022	Scaffold
B. waltoni	395577	GCA_014737395.1	24/09/2020	Scaffold

Identification of endogenous viral elements

Our six-step workflow to identify putative NIRVS commenced with an initial screening phase on the Galaxy platform (https://usegalaxy.org/). This first step involved Open Reading Frame (ORF) prediction using Getorf (v5.0.0.1, EMBOSS), targeting ORFs of 100–6000 nucleotides defined by start/stop codons [42]. The resulting ORFs were then subjected to a homology search against the viral protein database using Diamond (v2.1.11 + galaxy0) in Blastx mode. For this alignment, we applied a maximum E-value cutoff of 1 × 10 − 10 and considered up to 5 database matches per query sequence [43].

To refine the diamond output, we employed in-house Python scripts to select the optimal alignment hit for each query, prioritizing the lowest E-value and the highest bit score. Scripts are available in the following GitHub repository: < https://github.com/Bioinscripts/Pipeline-to-Endogenous-Viral-Elements-EVEs- >. The sequences were then taxonomically classified by querying the NCBI Entrez system. This classification guided a final filtering step designed to isolate RNA virus-like sequences by specifically removing any hits corresponding to retroviruses, DNA viruses, transposable elements, and other non-viral organisms manually.

To reduce redundancy amongst the identified sequences arising from paralogous insertions (copies), we performed a CD-HIT [44] run with a similarity threshold of 0.9 to cluster single sequences with their respective copies. Thus, we could reduce the amount of data for the next step, in which we aligned only the single sequences (or non-redundant) against the nr and core_nt databases (release gb263.0) on the online Blast platform (https://blast.ncbi.nlm.nih.gov/Blast.cgi). Amongst the 10 best hits in each analysis on Blastx, we selected the first viral hit (based on the lowest E-value) as main evidence to catalogue the query sequence as a putative NIRVS (Additional File 1).

Assessment of viral diversity, conserved domains and gc percentage

Taxonomic classification of the putative NIRVS was determined based on the NCBI taxonomic lineage of their respective best Blastx hits, referencing the NCBI Taxonomy Browser (https://www.ncbi.nlm.nih.gov/Taxonomy/Browser/wwwtax.cgi). For conserved domain assessment, ORFs and their corresponding amino acid sequences were first predicted from the endogenous viral sequences using NCBI's ORFfinder (https://www.ncbi.nlm.nih.gov/orffinder). These amino acid sequences were subsequently analyzed for conserved domains using both NCBI's Conserved Domain Database (CD-Search) [45] and InterProScan [46]. To measure GC content on viral sequences, we used geecee (Galaxy Version 5.0.0) with default parameters [42].

Retrieval of circulating viruses in bees and alignment with viral elements

To assemble a comprehensive dataset of viral sequences associated with arthropod hosts, we conducted a systematic search of the NCBI Nucleotide database. Recognizing the variability in host annotations across records, we employed a broad search strategy incorporating taxonomic terms such as "Arthropoda," "Insecta," and specific orders including "Hymenoptera," "Diptera," "Lepidoptera," "Coleoptera," "Hemiptera," "Orthoptera," "Odonata," "Blattodea," "Mantodea," "Araneae," "Scorpiones," "Acari," and "Decapoda." Additionally, we included terms related to specific genera and groups of interest, such as "Apoidea," "Apis," "Bombus," "Melipona," and "Euglossini," to capture sequences pertinent to bees and related taxa. This approach aimed to capture a wide array of arthropod-associated viral sequences, accounting for potential inconsistencies in metadata annotations. All retrieved sequences were downloaded in FASTA format for subsequent analyses. This comparison aimed to assess the alignment of 15 contiguous bases between the EVEs and known viruses, based on the literature to predict a possible silencing function [47].

Assessing shared EVEs between different Bombus species

To assess the similarity of the NIRVS shared among different species of the Bombus genus, we used BLASTn alignment data. The identifiers of each alignment pair were sorted and combined, ensuring that each unique pair was counted only once. The NIRVS pairs were subsequently enumerated by species, thereby yielding a count matrix for each species pair. These counts were transformed into a matrix and standardized for dissimilarity analysis using the Bray–Curtis distance, implemented via the vegdist function from the vegan package (v2.6–10) [48]. The graph was made by the ComplexHeatmap package (v2.22.0). This methodological approach facilitated the identification of patterns of NIRVS sharing across the analyzed species, thereby highlighting potential evolutionary relationships or conserved functional roles of these genetic elements. It is noteworthy that stringent filtering procedures were not implemented during the BLASTn alignments (e.g., by employing identity or coverage thresholds). This methodological decision enabled the observation of raw interaction patterns and distributional behavior of EVEs across species.

de novo transcriptome assembly and EVE-derived transcriptional activity

A total of thirty RNA-seq libraries, one per unique species (Additional File 1) were retrieved using the iseq tool v1.5.0 [49]. Quality assessment of raw reads was performed with MultiQC v1.28 [50], followed by adapter trimming and removal of low-quality bases (Phred score < 20) using Fastp v0.24.0 [51]. High-quality paired-end reads were de novo assembled with SPAdes v4.1.0 in RNA mode [52]. Redundant transcripts were clustered at 95% sequence identity using CD-HIT-EST v4.8.1[44]. ORFs were predicted with TIdeS v1.3.5 [53], employing a custom-trained machine learning model based on a non-redundant Bombus genus proteome obtained from NCBI. Transcript abundance was quantified at the transcript level with Salmon v1.10.3 [54] and normalized by transcripts per million (TPM).

Mapping small RNA libraries and piRNA analysis

To investigate the potential production of piRNAs by Bombus species targeting the characterized NIRVS, we analyzed publicly available small RNA sequencing (sRNA-seq) libraries retrieved from the NCBI Sequence Read Archive (SRA). All datasets available up to January 10th, 2026 were considered, with the exception of Bombus terrestris, whose piRNA pathway has been previously characterized and described in detail [55]. The libraries analyzed, and their associated metadata, are provided in Additional File 1.

Raw sequencing reads were initially subjected to quality control and adapter trimming using fastp v1.0.1 [56]. Only high-quality reads passing the default quality filtering parameters were retained for downstream analyses. Filtered reads were subsequently mapped against the NIRVS sequences using Bowtie v1.3.1 [57], allowing up to one mismatch and retaining only valid alignments.

To evaluate piRNA-associated features, the size distribution of mapped reads was assessed, with particular emphasis on the 24–32 nt length range. Additionally, nucleotide composition analyses were performed to examine the characteristic 5′ uridine (1U) bias and adenine enrichment at the tenth position (10A), which are hallmark signatures of primary and secondary piRNA populations [58]. Read length profiles and nucleotide preference plots were generated using ggplot2 in R.

NIRVS GC context

For the analysis of NIRVS in Bombus terrestris, we further refined our selection based on GC content profiles across the flanking regions. We prioritized sequences that exhibited abrupt inflection points in GC content at the NIRVS insertion site. Each selected region comprised 20,000 nucleotides—10,000 upstream and 10,000 downstream of the NIRVS. To aid in this selection, we used the EMBOSS tool cpgplot to identify and visualize CpG islands across the extracted sequences (cpgplot -sequence extracted_sequences.fasta -graph ps -outfile cpgplot_results.txt) using the following parameters: window size of 500, minimum island length of 200, observed/expected CpG ratio threshold of 0.3, and minimum GC percentage of 30%.

These regions were then examined for their genomic context, distinguishing whether the NIRVS was located within exons, introns, or intergenic regions. This enabled more targeted analyses, including specific alignments between the NIRVS and the exon into which it was inserted.

Phylogenetic analysis

For this step, our goal was to assess the phylogenetic relationships among all 41 species in this work to infer how the shared EVEs relate to one another. We based our study on the most comprehensive phylogeny for the Bombus genus found in the literature [59]. We recreated their phylogenetic tree using the same sequences as the authors and included species not present in the original tree but available in GenBank. Sequences were edited and aligned (default parameters, CLUSTAL W) in MEGA v12 [60]. We applied the same logic to manual editing of the sequences; however, we excluded PEPCK (Phosphoenolpyruvate Carboxykinase) sequences from our analysis due to high alignment ambiguity. We selected the best substitution models in ModelFinder [61]. The tree was estimated in IQ-TREE v2.4.0 [62] with default parameters and up to 1000 bootstrap replicates.

We also estimated phylogenetic trees for the order Martellivirales, including predicted endogenous viral elements specific to the taxon. We retrieved all Martellivirales RdRp protein sequences from RefSeq (Search string: "Martellivirales[Organism] AND RNA-dependent RNA polymerase[Protein Name] OR RdRp[Protein Name] OR replicase[Protein Name] OR RNA dependent RNA polymerase[Protein Name] OR replication-associated polyprotein[Protein Name] OR replicase protein[Protein Name] OR non-structural polyprotein[Protein Name] OR nonstructural polyprotein[Protein Name]") and one sequence from each family in Hepelivirales and Tymovirales for the outgroup. Before aligning our sequences, we used the latest ICTV Taxonomy Release (< https://ictv.global/msl/current >) to classify the retrieved sequences into families and orders, filtering out all sequences that did not belong to Martellivirales but were collected with the search string, and later used the family classification to correctly address the monophyletic clades in the final tree. Using Mafft v7 [63] with the FFT-NS-I strategy, the amino acid sequences were aligned with the Xiangshan martelli-like Virus 1, Xiangshan martelli-like virus 2, and Xiangshan martelli-like virus 3 RdRp sequences, alongside all NIRVs that were similar and aligned to the same region in Xiangshan martelli-like viruses with blastx alignments. After manual trimming in MEGA v12, which led to shortening the sequences to an extension of 161 amino acids due to the short size of NIRVS, the tree was inferred using IQ-TREE Galaxy Version 2.4.0 + galaxy1 [62] with the Q.pfam + F + I + R4 substitution model, selected as the best-fit model by ModelFinder [61], and visualized with iTOL [64]. A thorough optimization strategy was employed, including 200 initial trees, 50 retained top trees, and an exhaustive Nearest Neighbor Interchange search. We assessed branch support using 1000 replicates of both SH-aLRT and Ultrafast Bootstrap. The final command line with personalized parameters is as follows: “iqtree –prefix PREF -T ${GALAXY_SLOTS:−10} –redo -s '/mnt/pulsar/files/staging/13893363/inputs/dataset_6ebff454-1cb5-4c94-ae97-e2af9d62d8f1.dat' –seqtype AA –lmap '1000' -m '' –msub 'nuclear' –cmin '2' –cmax '10' –merit 'AIC' –ninit '200' –ntop '50' –nbest '5' –nstop '100' –radius '6' –perturb '0.5' -allnni –alrt '1000' –sup-min '0.0' –ufboot '1000' –nmax '1000' –nstep '100' –bcor '0.99' –beps '0.5'”.

Genomic context characterization

We characterized the genomic context of all eight annotated Bombus species available in NCBI (B. affinis, B. bifarius, B. flavifrons, B. huntii, B. pascuorum, B. pyrosoma, B. terrestris, and B. vancouverensis) in order to find common patterns of integration or shared context among species, as well as proximity to transposable elements. To evaluate the identity and coverage shared among all flanking regions, NIRVS loci were first aggregated into clusters based on genomic proximity, merging intervals located within 10 kb of one another, and then extracted and concatenated regions spanning 30 kb upstream and downstream of each NIRVS cluster. Structural similarity was assessed using ProgressiveMauve [65], generating percent identity matrices that were clustered in R using the Ward.D2 method. Sequence similarity was evaluated with BLASTn, retaining the best high-scoring pair (HSP) per comparison. Mean nucleotide identity and cumulative coverage were calculated, and results were visualized as bubble plots using ggplot2, integrating Mauve clustering order with BLAST similarity.

Transposable Elements were annotated using a custom automated pipeline integrating de novo discovery with homology-based detection. Repetitive consensus sequences were identified using RepeatModeler2 [66] and merged with the Dfam database. To minimize redundancy, the combined repetitive consensus and Dfam database were clustered using cd-hit-est (version 4.8.1) [44] with a sequence identity threshold of 98% (-c 0.98) and alignment coverage of 80% (-aS 0.80). The resulting non-redundant library served as the reference for genomic masking and annotation using RepeatMasker version 4.2.2 [67]. To evaluate the spatial association between NIRVS and transposable elements (TEs), TE annotations were filtered to exclude non-TE features (e.g., rRNA, simple repeats) and merged if located within 1 kb. We calculated TE base-pair density within strand-corrected 500-bp bins extending 10 kb upstream and downstream of NIRVS insertion sites. Enrichment relative to the genomic background was assessed using Fisher’s exact tests.

Finally, to compare and investigate the genomic neighborhoods of NIRVS insertions across different Bombus species, we used the same window extending 30 kb upstream and downstream of the insertion boundaries as before, from which we mapped cellular genes using the GFF3 annotations available in NCBI, and repetitive elements using the RepeatMasker outputs, filtering out simple repeats and low-complexity regions. Transposable elements were color-coded by classification (LTR, LINE, SINE, DNA, RC) and visualized alongside EVEs and host genes using the matplotlib package.

Features composition comparison

In this analysis, we focused exclusively on viral coding sequences previously identified in Bombus spp. and on NIRVS found in Bombus terrestris. To reduce redundancy, the viral CDSs were filtered using CD-HIT with a 90% similarity threshold. Subsequently, we selected those viral CDSs that clustered with NIRVS, prioritizing sequences likely to share similar genomic features.

To investigate patterns of similarity between NIRVS and exogenous viral sequences, we performed hierarchical clustering based on codon usage, GC content, and sequence length. Codon usage profiles were generated using the cusp tool from EMBOSS, and GC content was calculated using geecee [42]. To identify the variables most strongly contributing to the observed clustering, we conducted a principal component analysis (PCA). In addition, to assess the possibility of using these features to differentiate between exogenous and endogenous viral sequences, we conducted a t-distributed stochastic neighbor embedding (t-SNE) using 2 dimensions and 6 perplexity [68].

PCR and sanger sequencing

We collected 3 Bombus spp. bees from wild colonies near Montana State University in Bozeman, Montana, USA. DNA was extracted following a phenol–chloroform-isoamyl alcohol protocol. 4 pairs of primers for PCR were designed from the best-representative EVEs for a broader range of Bombus species, based on blastn alignments (additional file 2, Fig. 10A) against Dumyat virus and Xiangshan martelli-like virus 2. After running the PCR products on an electrophoresis gel, we purified the amplified DNA sequences and Sanger-sequenced them. A final blastn alignment was performed to confirm the similarity between the sequenced element and the identified EVE.

Results

NIRVS are widespread in Bombus genomes

The genus Bombus (bumble bees) is well represented in GenBank, with numerous high-quality sequenced genomes, many of which are assembled to the chromosome level. Key contributions to this resource have been made by institutions such as the Chinese Academy of Agricultural Sciences and the Wellcome Sanger Institute. Bombus genome sizes exhibit considerable variation, ranging from approximately 230 Mb (Bombus superbus) to 479 Mb (Bombus vosnesenskii) (Fig. 1A). While the number of predicted ORFs generally correlates with genome size, the subsequent counts of Diamond Blastx hits and the final set of putative NIRVS (Additional File 2 – Fig. 1) do not maintain this direct proportionality. The marked reduction in candidate elements through these successive analytical stages underscores the critical role of stringent filtering in accurately identifying NIRVS.

Fig. 1.

Overview of the genomic features and endogenous viral elements identified in Bombus bees. A multiple bar graph exploring genome size, number of predicted ORFs, filtered DIAMOND viral hits, and putative EVEs after manual curation for each bee species. On the left, species are associated with their respective subgenera classification: Al, Alpinobombus; Ag, Alpigenobombus; Bi, Bombias; Bo, Bombus; Cu, Cullumanobombus; Kl, Kallobombus; Md, Mendacibombus; Mg, Megabombus; Ml, Melanobombus; Or, Orientalibombus; Pr, Pyrobombus; Ps, Psithyrus; Sb, Sibiricobombus; St, Subterraneobombus; Th, Thoracobombus. B Sankey plot displaying the NIRVS classification based on sequence similarity best hit across different taxa and genetic types

Overall, our comprehensive analysis of publicly available genome assemblies from 41 Bombus species identified 467 putative NIRVS sequences. These elements, identified via Blastx searches and subsequent filtering, ranged from 100 to 6000 bp in length, with 252 (approximately 54%) being shorter than 500 bp (details in Additional File 1). The distribution of NIRVS varied significantly among species: for instance, B. terrestris harbored 37 NIRVS, whereas B. hortorum and B. opulentus each contained only a single identified NIRVS. Notably, B. waltoni and B. superbus were the only species in which no NIRVS were detected under our analytical conditions.

Dominance of unclassified and plant/insect/fungi viral signatures in Bombus NIRVS

A substantial portion of the identified NIRVS (231 sequences) remained unassigned to any specific viral family or order. Among the classifiable sequences, NIRVS showed similarities with several viral families: Partitiviridae (32 sequences), Virgaviridae (n = 29), Phasmaviridae (n = 25), Totiviridae (n = 15), and Rhabdoviridae (n = 10). Additionally, some NIRVS could only be resolved to the order level, including 4 sequences assigned to Bunyavirales and 121 to Martellivirales. Consequently, the majority of these classified NIRVS likely originated from positive-sense single-stranded RNA (+ ssRNA) viruses, followed by those from double-stranded RNA (dsRNA) and negative-sense single-stranded RNA (-ssRNA) viruses (Fig. 1B).

Polymerase-related domains are prevalent in Bombus NIRVS

To confirm or refute the protein domains identified by Blastx alignments, we performed functional annotation of conserved viral domains within the identified NIRVS, revealing a prevalence of RNA-dependent RNA polymerase (RdRp) domains, detected in 66 NIRVS. Other common domains included Polymerase-related ones (methyltransferase: 20; helicase: 20) and coat protein domains (9 NIRVS) (Fig. 2A). However, a considerable number of sequences (292 NIRVS) lacked any detectable conserved viral domains, potentially indicating they are highly degraded or divergent viral remnants. Also, most species had more abundant RdRp integrations than in other domains, both in Blastx and CD-blast hits (Fig. 2B and C). Interestingly, while nucleotide-level alignments for many full-length NIRVS were not conclusive in establishing clear viral resemblance, the protein domains identified through functional annotation consistently corresponded to the viral protein families suggested by the initial Blastx alignments. This was particularly evident for NIRVS related to Xiangshan martelli-like virus 2 and Xiangshan martelli-like virus 3, from which most functional annotations originated. Many NIRVS contained multiple domains attributed to these specific martelli-like viruses. For example, in Bombus affinis, Bombus flavifrons, and Bombus terrestris, individual NIRVS related to these viruses were found to harbor up to three distinct conserved domains.

Fig. 2.

Screening of conserved viral domains in NIRVS. A Compared prevalence of predicted conserved domains (Interpro and CD-Blast) and protein hits on Blastx. B Shared Blastx protein hits between NIRVS across bee species. C Shared conserved domains between NIRVS across bee species. D multiple bar graph for conserved domains distribution across bee species and identified by related viral taxon

We observed a distinct profile of viral sequences integration associated with each taxon. NIRVS associated with Partitiviridae are typically derived mostly from polymerase sequences containing RdRp domains. In contrast, those linked to Phasmaviridae, Rhabdoviridae, Totiviridae, and unclassified Bunyavirales were mostly associated with nucleoprotein and coat/capsid and often lacked detectable conserved domains. Furthermore, some Virgaviridae-like NIRVS possessed both helicase and RdRp domains, which supports a broader pattern of diverse integrations from the order Martellivirales (see Fig. 2D and Additional File 2 – Fig. 2).

NIRVS size correlates with presence of conserved domains and host GC similarity

We separately analyzed the correlation between protein domains and two key genomic features of NIRVS sequences. To provide evidence on the origin and timing of integration, we analyzed GC content in the viral sequences. We compared them with the host sequences, where the GC content differences (Host GC – EVE GC = ΔGC) varied across viral families, in which NIRVS related to taxa such as Totiviridae, Martellivirales, Virgaviridae (including positive-sense ssRNA and dsRNA virus groups), and most of the unclassified sequences exhibited greater ΔGC values (Fig. 3). On the other hand, NIRVS related to Phasmaviridae, Rhabdoviridae, and Bunyavirales (all negative-sense ssRNA virus groups) showed more similarity to their hosts’ GC content.

Fig. 3.

Relationship between GC content and presence of conserved domains. Each NIRVS classified by Blastx hit was screened for the confirmation of conserved domains. Absence of predicted conserved domains for the blastx hits is labeled as green. Presence of predicted conserved domains with Interpro and CD-Blast for the blastx hits is labeled blue. Significance between GC content of NIRVS with and without domains was calculated using Wilcoxon test

Statistical comparisons revealed a significant tendency for sequences with functional annotation to show fewer minor differences from their host GC (p = 0.0037), suggesting that larger viral insertions with identifiable conserved domains may be more structurally adapted to the host genome (Fig. 3). Some domain-containing sequences, especially RdRp and coat proteins (p < 2e-16 and p = 0.00092, respectively), have significantly lower ΔGC compared to their counterparts. In contrast, sequences lacking conserved domains showed no significant correlation with genome size (R = 0.054, p = 0.57) and exhibited higher ΔGC values. This could reflect a selective pressure favoring GC content convergence toward host genomes for functional EVEs.

In terms of sequence length, the other genomic feature we highlight here, annotated sequences tend to be longer than non-annotated sequences in some cases. NIRVS with RdRp and polyprotein domains were significantly longer (p = 4.3e-12 and p = 0.00043, respectively), whilst no significant differences were observed for other virus-related proteins (Additional File 2—Fig. 3). NIRVS related to Virgaviridae and Totiviridae families generally had longer sequences, while those associated to sequences from Partitiviridae and Phasmaviridae families exhibited shorter average lengths.

NIRVS sharing patterns and shared genomic context suggest a high prevalence of inheritance events

To determine whether NIRVS were shared across Bombus species, we performed a Bray–Curtis dissimilarity analysis on all identified NIRVS sequences, which resulted in their clustering into five distinct groups (Fig. 4). When compared with results from the Bombus phylogeny, these groupings suggest potential shared histories of viral integration events among different subgenera. Our species are grouped according to previous phylogenies [59, 69–71]. The complete version of the tree is found in Additional File 2 – Fig. 4.

Fig. 4.

Phylogeny correlates with NIRVS sharing patterns in Bombus species. NIRVS were clustered using Bray–Curtis dissimilarity, which quantifies differences in sequence abundance between species, and visualized as a heatmap. These patterns were compared against a reduced phylogeny of 41 Bombus species inferred from four genes (16S, EF1-α, Opsin, and ArgK) using a maximum likelihood approach with 1,000 bootstrap replicates. The full phylogeny, comprising > 200 species, is provided in Additional File 2 – Fig. 4. Species are associated with their respective subgenera classification and colored accordingly: Al, Alpinobombus; Ag, Alpigenobombus; Bi, Bombias; Bo, Bombus; Cu, Cullumanobombus; Kl, Kallobombus; Md, Mendacibombus; Mg, Megabombus; Ml, Melanobombus; Or, Orientalibombus; Pr, Pyrobombus; Ps, Psithyrus; Sb, Sibiricobombus; St, Subterraneobombus; Th, Thoracobombus

Alpinobombus (Al), Bombus (Bo), and Pyrobombus (Pr) bees have the most identified NIRVS among all groups and are relatively close in the phylogenetic tree. This specific lineage may have conserved more endogenous viral sequences over time, as opposed to recent integrations, since, even though they do not share some sequences with other branches, they do share “Dumyat virus”, “Wuhan insect virus 22”, “Xiangshan martelli-like virus 2”, and “Xiangshan martelli-like virus 3” related EVEs with many other species.

From the total, only 6 NIRVS were species-specific, highlighting the high prevalence of NIRVS sharing and possibly inheritance. On the other hand, for some of the NIRVS, for instance, we observed the presence of Dumyat virus-related sequences in 38 out of 41 species, which correlated with the speciation of subgenus Mendacibombus (Additional File 2 – Fig. 4). Genomic context analysis of species with available annotations revealed a high level of nucleotide conservation and sharing for orthologous insertions of Dumyat virus and Xiangshan Martelli-like viruses (Additional file 2 – Figs. 5, 6, 7 and 8). Those regions display high microsynteny and conserved flanking genes, suggesting an orthologous insertion event rather than multiple insertions. They also appear to be inserted within clusters of diverse transposable elements (DNA, LINE, LTR), while the NIRVS locus itself remains relatively low in TE insertions (Additional file 2 – Fig. 9).

Many NIRVS are transcriptionally active in Bombus

To investigate the potential transcriptional activity and assess the distribution of the analyzed non-retroviral integrated RNA virus sequences within the Bombus genus, we mapped reads from each retrieved RNA-seq library against a set of 467 NIRVS sequences alongside constitutive host marker genes.

From the total, 83 (~ 20%) NIRVS showed transcriptional activity in at least one Bombus species. Interestingly, B. bifarius EVE5 was the most widespread, exhibiting notable transcript abundance in 20 distinct Bombus libraries derived from various species. Excluding NIRVS that showed no detectable transcript levels, the least widespread elements included B. terrestris EVE22, B. rufofasciatus EVE2, and B. picipes EVE15 (Fig. 5 and Additional File 1).

Fig. 5.

Assessment of Transcriptional activity of NIRVS sequences in Bombus species. Heatmap showing NIRVS abundance along different Bombus species. Rows were clustered based on Pearson correlation to group sequences with similar abundance calculated as transcripts per million (TPM). Only transcriptionally active sequences are shown

Notably, several NIRVS displayed high expression levels. Specifically, B. breviceps EVE3, B. terrestris EVE7, and B. impatiens EVE1 exhibited the highest Transcripts Per Million (TPM) values. In some libraries, the TPM levels for these NIRVS were comparable to those observed for constitutively expressed genes such as Histone H4 and Cathepsin L. Some sequences appeared to be expressed in different species due to the similarities that those sequences share across different species.

In order to check the fidelity of our findings, we assembled the reads from each RNAseq library that mapped against each EVE and performed blastn alignments to confirm that the assembled contigs match the integrations in size and identity (additional file 2, Fig. 13).

Potential for Bombus NIRVS to template piRNAs targeting exogenous viruses

Among the identified NIRVS in Bombus, 11 exhibited a minimum of 15 nucleotides (nt) of perfect contiguous identity when aligned against Arthropoda-associated viral sequences from the NCBI RefSeq database (Fig. 6). This observation is notable because PIWI-clade Argonaute proteins utilize piRNAs as guides to combat viral infections, among other functions [72]. Effective PIWI-catalyzed target slicing requires piRNAs to have at least 15 nt of contiguous base-pairing with the target sequence, while mismatches at other positions can be tolerated [47].

Fig. 6.

Similarity of NIRVS and circulating viruses in Bombus. Sequences are displayed according to nucleotide alignment between putative NIRVS and viruses from RefSeq database. Identity and sequence lengths are shown. Only sequences with at least 15 contiguous base pairs were considered

Therefore, even though Bombus NIRVS generally show low overall nucleotide identity and coverage when compared to contemporary circulating viruses, these identified 15nt contiguous regions could hypothetically allow NIRVS-derived transcripts to serve as templates for piRNA biogenesis, thereby potentially contributing to antiviral defense mechanisms.

We also tested our hypothesis by mapping small RNA reads from public libraries against the identified NIRVS. We did not find many matches for most EVEs, and even when matched, there were very few reads that could not be confidently classified as piRNAs (Additional File 2 – Figs. 14 and 15).

NIRVS and exogenous viruses show similar codon usage and GC content

To investigate whether NIRVS maintained viral molecular signatures or converged to the host characteristics, we used Bombus terrestris as a model to assess nucleotide composition that could indicate adaptation and help to differentiate exogenous from endogenous sequences. Through a Principal Component Analysis (PCA) based on codon usage, GC content, and sequence length, we observed that a few codons and, most notably, GC content were key factors shaping the distribution of sequences in multivariate space (Additional File 2 – Fig. 11). Also, NIRVS encoding conserved protein domains tended to cluster apart from those lacking annotated domains, suggesting potential functional constraints associated with recognizable protein-coding capacity.

We also tried a different strategy, trying to differentiate endogenous and exogenous sequences. Comparison among NIRVS and related exogenous viruses identified in Bombus terrestris revealed that for some NIRVS the patterns of codon usage closely match those of related circulating viruses, which suggests two possibilities: (i) NIRVS are products of recent integrations or (ii) NIRVS are under positive selection, preserving exogenous viral characteristics (Additional File 2—Fig. 12).

We also compared NIRVS in Bombus terrestris with similar viral elements from other species and exogenous viruses, all related to the order Martellivirales, through a phylogenetic analysis (Fig. 7). We aimed to identify which lineage within this taxon is most closely related to the NIRVS we previously identified and how the sequences are related to one another. All viral elements were grouped into a highly supported monophyletic clade (SH-alrt = 98.2/Bootstrap = 99), suggesting a shared evolutionary event or a specific viral lineage that diversified within these bee hosts, rather than random, unrelated insertions. They are placed as a sister clade to Xiangshan martelli-like virus 2 and Xiangshan martelli-like virus 3, strongly suggesting that they constitute a distinct lineage within the broader Martellivirales order.

Fig. 7.

Maximum likelihood phylogenetic tree (partial RdRp amino acid sequences) of NIRVS identified in Bombus genomes. The phylogenetic reconstruction places the identified NIRVS within the order Martellivirales. The tree was inferred using IQ-TREE (Galaxy Version 2.4.0 + galaxy1) with the Q.pfam + F + I + R4 substitution model, selected as the best-fit model by ModelFinder, and visualized with iTOL v7. A thorough optimization strategy was employed, including 200 initial trees, 50 retained top trees, and an exhaustive Nearest Neighbor Interchange (-allnni) search. Nodal support values are indicated as Shimodaira–Hasegawa—approximate likelihood ratio test (%)/Ultrafast Bootstrap (%) based on 1000 replicates for both. The EVEs (colored in red) form a well-supported monophyletic clade (98.2/99) and are positioned as the sister group to the Xiangshan martelli-like viruses 2 and 3. Unclassified Martelli-viruses are depicted as single branches rather than collapsed within families. The complete version of this tree, without any collapsed branches, is shown in Additional File 2, Fig. 16

Discussion

The integration of non-retroviral RNA viruses into bumblebee genomes represents a dynamic evolutionary process that elicits both past viral infections and bee-virus coevolution. In this study, we characterized NIRVS across several Bombus genomes, revealing significant variation in their sequence composition, functional domains, and taxonomic origins. Our findings suggest that viral integrations have occurred repeatedly throughout the evolutionary history of bumblebees, and many are shared among Bombus species, likely due to inheritance from a common ancestor. The PCR results further validate our findings and show the actual presence of the two most widespread NIRVS across Bombus species (Additional File 2, Fig. 10B, C and D).

The most frequently found RNA virus families related to EVEs in arthropod genomes are primarily four: Rhabdoviridae, Flaviviridae, and Chuviridae [2, 55, 73]. The last two weren’t identified in our analysis. The viral elements identified showed similarities to virus sequences classified within families that do not have significant pathogenic importance in bees, since the main viruses affecting commercially important populations belong to the families Dicistroviridae and Iflaviridae included in the order Picornavirales [74, 75]. However, the origin of these NIRVS from viral groups with currently limited recognized pathogenicity in Bombus does not necessarily negate their biological relevance. Indeed, endogenous viral elements, regardless of the pathogenic potential of their ancestral viruses, are increasingly understood to contribute to host biology, including the potential modulation of antiviral immune responses.

When assessing the potential immune role of NIRVS in Bombus species, we searched for exogenous viruses that shared a 15nt contiguous sequence identical to intrinsic regions in integrated elements. The choice for this criterion is based on how the piRNA pathway functions in antiviral immunity, providing initial evidence of a possible role for these integrations [58]. We included sRNA sequencing analysis to further understand this possibility, but there were very few publicly available libraries, and our results were inconclusive. We only detected a small portion of small RNAs mapping to PASCUORUM_EVE2 from library SRR29032815. The absolute number of reads was very modest (less than 20), and even though their molecular profile displayed some canonical hallmarks of primary piRNAs (such as the size of 27–29 nt and a strong uracil bias at the 5' position), they lacked ping-pong signatures, which are robust hallmarks for secondary piRNAs. The absence of secondary piRNAs might be related to the inexistence of the target transcript, which would enable the ping-pong cycle. We still consider the possibility that secondary piRNA reads could be underrepresented due to the small amount of germline tissue compared to other tissues in the whole bees, since all samples for the screened libraries were from whole individuals.

Apart from antiviral immunity, EVEs may be conserved or exaptated to perform different roles within the cell [2]. Many of our findings suggest that the integrations do not retain fully conserved protein domains or do not show similarity with known viral domains. Also, some viral lineages frequently undergo modular gene exchange. Such events can produce genomes composed of domains with distinct evolutionary origins, which in turn affects how individual modules are retained or lost during endogenization. The presence of EVEs encoding only capsid genes, or only replicase domains, could reflect ancestral recombination events.

The virus families Partitiviridae and Totiviridae (dsRNA viruses), along with Virgaviridae (a + ssRNA virus family within the order Martellivirales), are recognized for infecting a broad range of hosts, including fungi, plants, and protozoa [76–78]. The detection of NIRVS in these families across various insect species has been previously documented [79], consistent with our findings in Bombus.

The family Rhabdoviridae and the order Bunyavirales (which includes the family Phasmaviridae) consist primarily of -ssRNA viruses, many of which are known to infect insects [80]. NIRVS derived from these viral taxa are frequently reported in arthropods, where Rhabdoviridae-related integrations, in particular, appear to be notably abundant [55, 79, 81–83].

From an evolutionary perspective, the strong correspondence between NIRVS grouping and the Bombus evolutionary history suggests vertical inheritance of many viral sequences from common ancestors prior to species diversification. Groups in which we identified more sequences are probably under similar pressures to conserve those elements, or integrations are more likely to be recent. The same logic applies to groups that harbor few conserved viral sequences, which may have lost them due to evolutionary events. GC content, codon usage, transcription activity, and genomic context analyses were key factors in understanding the intricacies of NIRVS inheritance and conservation, from which we identified patterns of conservation and activity in viral sequences across the analyzed Bombus species. Recently integrated sequences tended to have GC% closer to that of their source virus, while older insertions tend to have GC% similar to that of their host. The same logic applies to codon usage: the endogenous viral sequence might undergo neutral mutations and purifying selection, approximating it to the host sequence nucleotide structure.

We hypothesized that the frequent interactions between plants and bees, and the presence of commensal organisms such as gut microbiota and parasites, might enable the integration of different viruses into bee genomes. Most recently, evidence of arthropod-infecting viruses belonging to those families, especially in dipterans and ticks, has been reported (reviewed by [84]). It indicates the need for further investigation of insect viromes to better understand their diversity and the mechanisms underlying cross-species transmission. However, it is known that insects may vector a variety of plant viruses, some of which promote propagative-persistent infections in which they can fully replicate in this intermediate host, even in germline cells (reviewed by [85]). It is important to consider that the ubiquitous presence of arthropods across global biomes and ecological niches, the high diversity of arthropod viruses, and the reliance on solely innate antiviral responses all contribute to the complexity of arthropod-virus interactions.

Non-retroviral EVEs in Bombus species were assessed in two previous works. The first [61] presented data regarding the presence of EVEs in the genomes of 48 arthropods, including A. mellifera and B. terrestris. Although their findings support the hypothesis of the present study, they reported a different number of EVEs [51] for B. terrestris than those found here [37]. This discrepancy may be explained by the use of a different genome assembly, different database versions, or differing selection criteria in the methodology. They selected EVEs smaller than 100 bp and used a less stringent e-value for alignment searches. Additionally, sequences similar to single-stranded DNA viral sequences were included. A second study also assessed the presence of endogenous viral elements in various insects but focused on virga/nege-like elements [86]. Their findings align with ours, considering the presence of Martellivirales-related endogenous elements and a similar presence of various conserved domains. Both works included only a few species of Bombus bees and had access to less robust libraries in comparison to our work, due to the frequent updates of all databases in GenBank up until now.

Our findings suggest that bumblebees may have historically interacted with a diverse array of novel or highly divergent RNA viruses. Understanding of the bumblebee-associated ancient virome is essential for finding new solutions for modern problems and unveils the limitations of existing reference databases in accurately capturing the full extent of viral diversity in ecologically important insects. However, our study has some limitations. For instance, the identification of NIRVS relies primarily on sequence homology searches into reference genomes (not always assembled in chromosomes) and lacks experimental validation. Only the most conserved regions are detectable at the protein level due to the high divergence of viral sequences. In addition, our methodology excludes DNA viruses, is limited to publicly available genomes and annotations, and does not include laboratory validation to assess potential immune-related functions.

Conclusion

This study revealed that NIRVS in Bombus genomes vary significantly in sequence composition, functional domains, and taxonomic origin, with a great number of elements shared across species, probably due to inheritance from common ancestors. Interestingly, the elements detected were related to viral groups with currently limited recognized pathogenicity in bumblebees. Nonetheless, the biological relevance of these NIRVS should not be underestimated. Endogenous viral elements are increasingly acknowledged for their potential roles in host genome evolution, immune regulation, and antiviral defense. Collectively, these findings emphasize that NIRVS are not merely remnants of past infections but may also be conserved in many species for potential genomic roles.

Abbreviations

LTR

Long Terminal Repeat

NIRVS

Non-retroviral Integrated RNA Virus Sequences

ORF

Open Reading Frame

PCA

Principal Component Analysis

piRNA

PIWI-interacting RNA

PIWI

P-element Induced WImpy testis

RdRp

RNA-dependent RNA polymerase

sRNA

Small RNA

Authors’ contributions

Conceptualization, E.R.G.R.A. and M.A.C.; methodology, L.B.d.A.C., J.P.N.S., L.Y.M.F. and G.V.P.R; formal analysis, L.B.d.A.C., J.P.N.S., L.Y.M.F., and G.V.P.R.; resources, E.R.G.R.A.; data curation, L.B.d.A.C., J.P.N.S., L.Y.M.F. and G.V.P.R.; writing—original draft preparation, L.B.d.A.C., J.P.N.S., L.Y.M.F., G.V.P.R. and E.R.G.R.A.; writing—review and editing, L.B.d.A.C, J.P.N.S., L.Y.M.F., G.V.P.R, and E.R.G.R.A.; visualization, L.B.d.A.C., J.P.N.S., G.V.P.R., and L.Y.M.F.; supervision, E.R.G.R.A. and M.A.C.; project administration, E.R.G.R.A.; funding acquisition, E.R.G.R.A. All authors have read and agreed to the published version of the manuscript.

Funding

This study was financed in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior—Brazil (CAPES)—Finance Code 001. E.R.G.R.A. is a researcher of the Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq).

Data availability

All datasets analyzed during the current study are available in GenBank, RefSeq and Read Archive (SRA) from the National Center of Biotechnological Information (NCBI). The accession codes are available in Additional File 1.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.