Introduction

The interspecific transmission of pathogens occurs when a pathogen is transmitted to a host previously not infected by that pathogen1,2,3, one possible outcome of which is spillover, in which the pathogen establishes a new maintenance host. Viral spillover is widely considered a threat to wild insect pollinators, due to the possibility of virus spillover from managed western honey bee (Apis mellifera) colonies. There has been much focus on proposed viral spillover from A. mellifera to bumblebees (genus Bombus), many of which are experiencing population declines with some species facing a high likelihood of extinction4,5. Spillover is a multifaceted phenomenon6 and a term that is used to refer to a variety of distinct scenarios. We use the terms exposure, dead-end spillover, and spillover with range expansion as defined in Fig. 1. Briefly, the first and most common scenario is exposure, which is the incidental presence of a virus in or on an organism due to the sharing of a common environment with the true host, and is unrelated to infection. If a virus exposed to a new host is capable of at least initiating infection, but cannot be sufficiently transmitted between the new hosts, then the result is dead-end spillover2 in which the virus will inevitably be eliminated from the new host population (Fig. 1).

Fig. 1: Conceptualisation of three hypothetical spillover scenarios using A. mellifera (Apis) and B. impatiens (Bombus) as an example.
figure 1

Based on other frameworks regarding viral spillover1,2,3,6,7,8. The Apis and Bombus silhouettes by Melissa Broussard were obtained from https://www.phylopic.org/ and adapted with permission under the license: https://creativecommons.org/licenses/by/4.0.

Full size image

To establish a new host, a virus must overcome a complex set of barriers, including molecular interactions, as well as host ecology and other biological factors. As a result, exposure and dead-end spillover are vastly more common than spillover with range expansion, as is seen in other animal systems7,8. Even viruses which can overcome important barriers to establishing a new host may still end up at dead-end spillover. For example, even influenza viruses that can complete replication and produce large quantities of virions in a new host can have insufficient virus transmission to establish a new maintenance host8. We reserve the term spillover with range expansion for the scenario in which the virus overcomes these barriers and can maintain itself in the new host population through transmission between the new hosts (Fig. 1). As it can maintain itself, the presence of the virus in the new host is not dependent on contact with viral reservoirs in the origin host.

A key feature of spillover with range expansion is heightened evolution in the pathogen9 and the rapid emergence of variants that are adapted to the new host (Fig. 1), whereas in the exposure and dead-end spillover scenarios, the virus sequence identity remains essentially identical to that of the origin host due to unsuccessful propagation and adaption of the virus. Although there is a large body of research proposing spillover between honey bees and various other arthropods10, spillover-related scenarios have been poorly distinguished. The scenario of concern discussed in the literature is spillover with the threat of range expansion, as the establishment of a new maintenance host poses the greatest threat to host populations (Fig. 1). However, most of the spillover evidence is based on the assessment of virus presence, which has limited capability to distinguish between exposure, dead-end spillover, and spillover with range expansion. As a result, it is likely that cases of exposure or dead-end spillover have been widely mischaracterized as spillover with range expansion. Correctly identifying the spillover-related scenarios occurring in bees is crucial, as they pose different degrees of risk to pollinators, with the risk increasing from exposure to dead-end spillover, to spillover with range expansion.

A relatively narrow range of viruses have been studied in the context of bee virus spillover, mainly the A. mellifera viruses deformed wing virus (DWV, Iflavirus aladeformis), black queen cell virus (BQCV, Triatovirus nigereginacellulae) and sacbrood virus (SBV, Iflavirus sacbroodi)10. A key example of proposed viral spillover with the greatest potential for range expansion is the spillover of DWV A or B master variants from A. mellifera to B. impatiens. Wild Bombus species are commonly RT-PCR positive for both master variants of DWV11, and DWV is able to complete its replication cycle in some members of this genus, at least in artificial experiment systems12, indicating at a minimum dead-end spillover and DWV replication can be detected in wild Bombus species13,14. There is currently no evidence for maintenance of DWV in Bombus without continued exposure. The main hypothesised route of spillover for DWV and other viruses, is through oral ingestion of viruses deposited on shared flowers, as supported by the RT-PCR detection of DWV on flowers visited by A. mellifera13 There is, however, also a case against DWV spillover with range expansion. Firstly, there is little evidence of any negative outcomes of proposed viral spillover leading to range expansion in wild pollinators10,15. Other studies suggest that successful DWV transmission between bumblebees is implausible in natural contexts12,16, as DWV transmission is highly correlated with the abundance and prevalence of the Varroa mite which does not parasitise Bombus species17 and DWV infection is highly ineffective via the oral-faecal route13,18,19. Spillover has been proposed for a variety of other honey bee viruses, albeit with less available evidence than for DWV10.

Considering insights from other animal virus spillover systems7,8, the evidence for viral spillover from A. mellifera to other insects, appears to be highly uncertain. A genomic approach can address some of this uncertainty, specifically whether viral variants adapted to a new host have emerged following a spillover event. By comparing the genomes of the total viral community, variants that have emerged within a new host can be identified, which may indicate whether the presence of A. mellifera viruses in other insects is a result of exposure, dead-end spillover, or spillover with range expansion. To determine the state of viral spillover between A. mellifera and B. impatiens, we surveyed the meta-transcriptomes of sympatric A. mellifera and B. impatiens from the field study programme in Minnesota, USA. Using this RNA biased approach, we tested for evidence of spillover for all RNA viruses identified in both bees by attempting to identify B. impatiens-specific variants of A. mellifera viruses over a three-year period.

Results and discussion

Sequencing libraries were generated for a total of 408 Apis mellifera and 131 Bombus impatiens samples. For A. mellifera and B. impatiens, respectively, 19 and 14 sample libraries were excluded from the analyses because they generated no contigs or generated only contigs below 1000 bp, leaving a total of 389 and 117 samples with sequencing data. For A. mellifera and B. impatiens, respectively, the average number of reads (that passed the Q-score threshold) per library was 80,563 and 101,004, the average read length was 550 bp and 487 bp, and the percentage of reads that mapped to viral contigs was 9.7% and 4.6%.

Across all samples, viral communities were distinct between A. mellifera and B. impatiens (PERMANOVA Df = 1, R2 = 0.007, p = 0.001 ***). The only experimental variables which significantly impacted the community compositional similarity between A. mellifera and B. impatiens was the collection year (PERMANOVA Df = 2, R2 = 0.01, p = 0.001 ***), as it was the only variable with a significant interaction with the host genus (Supplementary Table 1; Supplementary Fig. 1).

Viral communities in B. impatiens were overall more homogenous between the sampling years than in A. mellifera. The key viruses that distinguished the two bee species were (assessed by PCA, Fig. 2, and Pearson correlation coefficients (>0.1), Supplementary Fig. 2, and community composition, Fig. 3), in A. mellifera: Iflavirus aladeformis (IA), apis rhabdovirus 1 (AR1), Cripavirus ropadi (CR), hubei partiti−like virus 34 (HPV34), Cripavirus mortiferum (CM), lake sinai virus 3 (LSV3), lake sinai virus 6 (LSV6), unclassified sinaivirus (US), Lake Sinai virus 2 (LSV2), Aparavirus apisacutum (AA), and Iflavirus sacbroodi (IS). B. impatiens was associated with vespa velutina associated permutotetra−like virus 1 (VVPV2), hymenopteran phasma−related virus OKIAV231 (HPVO231), hubei picorna−like virus 27 (HPV27), cyclosorus interruptus picorna−like virus (CIPV), ganda orthophasmavirus (GO), unclassified phasmaviridae (UP), mayfield virus 1 (MV1), hymenopteran phasma−related virus OKIAV234 (HPVO234), elf loch virus (ELV), bombus−associated virus reo1 (BVR1), agassiz rock virus (ARV), bactrocera tryoni iflavirus 1 (BTI1), bombus−associated virus Vir4 (BVV4), andrena haemorrhoa nege−like virus (AHNV), and allermuir hill virus 1 (AHV1). The remaining viruses had a neutral correlation (r = 0.0 or −0.0) with either host genus (Supplementary Fig. 2) and low relative abundances of reads (Fig. 3): Aparavirus israelense (AI), Lake Sinai virus 1 (LSV1), vespa velutina partiti−like virus 2 (VVP2), lake sinai virus (LSV), Triatovirus nigereginacellulae (TN), chronic bee paralysis virus (CBPV), and hymenopteran phasma−related virus OKIAV233 (HPVO233). Outside of a core set of viruses with well-established insect host ranges (DWV or IA, SBV or IS, BQCV or TN, BTI1, AHNV), the viruses presented by this study have uncertain host ranges, but have at least been identified by metagenomic studies in insects previously.

Fig. 2: Viral community similarity comparison between A. mellifera (Apis) and B. impatiens (Bombus).
figure 2

Unconstrained PCA plot showing only the top 23 viral assignments influencing community composition are shown as labels. This plot was based on CSS normalised read counts mapped against taxonomically classified contigs. AA Aparavirus apisacutum, AHNV andrena haemorrhoa nege-like virus, AHV1 allermuir hill virus 1, AR1 apis rhabdovirus 1, ARV agassiz rock virus, BTI1 bactrocera tryoni iflavirus 1, BVR1 bombus-associated virus reo1, BVV4 bombus-associated virus vir4, CM Cripavirus mortiferum, CR Cripavirus ropadi, GO ganda orthophasmavirus, HPV27 hubei picorna-like virus 27, HPV34 hubei partiti-like virus 34, HPVO231 hymenopteran phasma-related virus OKIAV231, HPVO233 hymenopteran phasma-related virus OKIAV233, IA Iflavirus aladeformis, LSV lake sinai virus, LSV1 Lake Sinai virus 1, LSV2 Lake Sinai virus 2, MV1 mayfield virus 1, US unclassified sinaivirus, VVAPV1 vespa velutina associated permutotetra-like virus 1, VVPV2 vespa velutina partiti-like virus 2.

Full size image
Fig. 3: Compositions of viral communities in A. mellifera (Apis) and B. impatiens (Bombus).
figure 3

Relative abundances of viruses in A. mellifera A–C and B. impatiens D–F in 2021, 2022, and 2023, respectively, and sample compositions also merged by apiary site (Crosby, Golf, and Vet). Relative abundances are based on CSS normalised read counts mapped against taxonomically classified contigs. AHV1 allermuir hill virus 1, AHNV andrena haemorrhoa nege-like virus, AA Aparavirus apisacutum (Acute bee paralysis virus, ABPV), AR1 apis rhabdovirus 1, BTI1 bactrocera tryoni iflavirus 1, BVV4 bombus-associated virus vir4, CM Cripavirus mortiferum, CR Cripavirus ropadi, GO ganda orthophasmavirus, HPV34 hubei partiti-like virus 34, IA Iflavirus aladeformis (Deformed wing virus, DWV), IS Iflavirus sacbroodi (Sacbrood virus, SBV), LSV lake sinai virus, LSV1 Lake Sinai virus 1, LSV2 Lake Sinai virus 2, MV1 mayfield virus 1, VVPV2 vespa velutina partiti-like virus 2.

Full size image

Across collection years, variation in viral community compositions led to minor differences in similarity between A. mellifera and B. impatiens. Communities in A. mellifera and B. impatiens showed slightly closer clustering in 2022 and 2023 than in 2021 (Supplementary Fig. 3). Overall, the viruses which contributed most to the A. mellifera and B. impatiens communities becoming more similar were LSV1, LSV2, US, LSV, GO, CM, UP, and HPVO233 (Supplementary Figs. 3 to 5). Viral community shifts were observed in the transplanted colonies, but overall, the communities remained similar to flower-collected bees in the same year (Supplementary Fig. 6A). One of the transplanted colonies showed a shift to have a greater proportion of B. impatiens-associated viruses, primarily MV1 (Supplementary Fig. 6B).

For all viruses identified in this study, the A. mellifera virus sequences in B. impatiens, and vice versa, did not form distinct clusters or clades indicating a strain resulting from spillover with range expansion (Fig. 4, Supplementary Fig. 7). Instead, they are likely the result of exposure or dead-end spillover (as defined in Fig. 1). This is consistent with other reports of RT-PCR amplicons in Bombus, which follow those in A. mellifera - for example, DWV-B that has come to be dominant in A. mellifera is also found in Bombus20 and it is unclear if host species has driven the emergence of A. mellifera virus strains specific to other arthropod hosts21. Our observations of highly similar A. mellifera viruses in B. impatiens is consistent with other spillover studies which report spillover only when Bombus and other arthropods are sympatric with A. mellifera10,21,22,23, which is more suggestive of exposure or dead-end spillover, since these viruses do not seem to be maintained in Bombus in the absence of A. mellifera.

Fig. 4: Similarity of viral sequences in A. mellifera (Apis) and B. impatiens (Bombus).
figure 4

Similarity network A of whole-contig average nucleotide identity percent, showing only edges over 98% identity. Maximum likelihood phylogeny B of RNA dependent RNA polymerase (Pfam PF00680). Tree constructed using amino acid sequence in IQTREE with 1000 bootstrap replications. Only bootstrap node values > 70 are shown. Data point key is for both (A, B). AHV1 allermuir hill virus 1, AHNV andrena haemorrhoa nege-like virus, AA Aparavirus apisacutum (Acute bee paralysis virus, ABPV), AI Aparavirus israelense (Israeli acute paralysis virus, IAPV), AR1 apis rhabdovirus 1, BTI1 bactrocera tryoni iflavirus 1, BVR1 bombus-associated virus Reo1, CM Cripavirus mortiferum, CR Cripavirus ropadi, GO ganda orthophasmavirus, IA Iflavirus aladeformis (Deformed wing virus, DWV), IS Iflavirus sacbroodi (Sacbrood virus, SBV), LSV lake sinai virus, LSV1 Lake Sinai virus 1, LSV2 Lake Sinai virus 2, MV1 mayfield virus 1.

Full size image

Overall, the occurrence of an A. mellifera virus in B. impatiens, and vice versa, coincided with shorter contig length (Supplementary Fig. 4, Supplementary Table 2). The contigs recovered were dominated by RNA viruses (Table 1), with the same key viruses (Fig. 2) showing the largest portion of recovered contigs and genomes (Table 1). Though comparable normalised read counts assigned to certain viruses being shared between A. mellifera and B. impatiens (Supplementary Fig. 5), the discrepancy in contig lengths between A. mellifera and B. impatiens suggested that the contigs resulting from apparent spillover generated more fragmented genomes. Exceptions to this were lake sinai virus, unclassified phasmaviridae, and Triatovirus nigereginacellulae, for which A. mellifera and B. impatiens did not have significantly different contig lengths (Supplementary Table 2). Across the three study years, the B. impatiens community maintained a consistent composition (Fig. 3), with the A. mellifera viruses never expanding to dominate the B. impatiens community. Of the A. mellifera pathogens identified, it was SBV or IS that formed the largest portion (M = 16%, SD = 10%) of the B. impatiens virome (Fig. 3) and full SBV genomes were frequently recovered from B. impatiens (Table 1, Supplementary Fig. 4). Despite this, SBV contributed little to the overall dissimilarity between the two communities (Fig. 2). Nevertheless, the sequence identity of these contigs remained at over 98% nucleotide identity between A. mellifera and B. impatiens (Fig. 4), suggesting that these high SBV levels in B. impatiens are simply the result of high exposure of SBV from A. mellifera, and not of spillover leading to range expansion within B. impatiens (Fig. 1). Together, the community composition and shorter contigs support the exposure or dead-end spillover scenario.

Table 1 Top abundance viral taxa with contig counts
Full size table

As a further examination of whether adaption might have occurred in the RNA dependent RNA polymerase (RdRp) gene, which is routinely targeted in RT-PCR surveillance programmes, we observed that no B. impatiens-specific clades of A. mellifera viruses could be identified (Fig. 4B, Supplementary Figs. 7- 8). Clustering based on the average nucleotide identity of whole contigs (Fig. 4A) and percent identity of RdRp gene sequences (Supplementary Fig. 9) often resulted in clusters composed of sequences derived from both A. mellifera and B. impatiens, with no clusters showing emerging subclusters or clusters specific to either host species. Reference sequences-based networks for sequences over 200 bp length from Apis mellifera, Apis cerana, and Bombus spp. only (Supplementary Figs. 10, 11) showed various strain clusters within Apis including A and B for DWV, and SBV strains in A. cerana, whilst networks based on sequences over 4000 bp from a variety of insect taxa showed similar strain distinctions (Fig. 5).

Fig. 5: Similarity networks of NCBI reference sequences of Iflavirus aladeformis (deformed wing virus, DWV), Iflavirus sacbroodi (sacbrood virus, SBV), and Triatovirus nigereginacellulae (black queen cell virus, BQCV) recovered from various insects.
figure 5

A Nnucleotide sequences over 4000 bp and B RNA dependant RNA polymerase protein sequences predicted from the nucleotide sequences in (A). Only edges over 98% identity are shown. The network in (A) is based on fastANI percent identity of whole contigs and in (B) it is based on BLAST (blastp) percent identity of RNA dependent polymerase (Pfam PF00680). Datapoints coloured by host origin of viral sequence as in key.

Full size image

In addition, the networks generated from public reference sequences showed a similar pattern - at 98% identity, DWV, SBV, and BQCV separate into distinct genotype clusters within A. mellifera, but form no Bombus-specific clusters (Supplementary Figs. 10, 11). For comparison, SBV separates into clear A. mellifera and Apis cerana strains24, which can be seen in our reference sequence networks, demonstrating that this network approach can separate virus strains adapted to different hosts (Supplementary Fig. 10, Fig. 5). There is also previous evidence of DWV divergence within A. mellifera itself (with DWV-B globally displacing DWV-A20) and of DWV showing genetic divergence when detected in different Apis species25. The Aparavirus complex in Apis26 also shows genetic adaptation in an insect virus over a short time scale. Furthermore, a recent study27 has detailed the shifted viral landscape in honey bees viruses across their geographical range, demonstrating how this network-based approach can reveal patterns of ongoing viral evolution within A. mellifera. In contrast, in hosts outside of Apis such as the bumblebees in this study, there is a lack of demonstrated evolutionary divergence of honey bee viruses, which reinforces the argument that the widespread detection of A. mellifera viruses in other insects may often be the result of exposure or dead-end spillover. There is, however, a need to expand upon the viral genomic datasets available for wild pollinators, as a clear strain cluster might only be observed after systematic surveillance has been conducted.

To date, the next-generation sequencing-based studies for the surveillance of pathogens in bees are still limited. Here we have shown that bumblebees have a distinct viral community, with the presence of A. mellifera viruses attributable to exposure or dead-end spillover from honey bees to bumblebees. The transplantation of B. impatiens colonies from indoors to sites with A. mellifera did not increase the proportion of A. mellifera viruses in B. impatiens (Supplementary Fig. 6). Instead, the variation in exposure to A. mellifera only seemed to produce a shift in the putative B. impatiens viruses already present in the colonies. This concurs with previous studies that has shown that Bombus colonies transplanted to sites free of A. mellifera eliminate DWV and BQCV and reestablish a Bombus-specific virome28, and that wild bee species including Bombus possess distinct viral communities, even when they are closely related and share a common environment29,30,31,32,33. Although we attempted to sample B. impatiens at varying levels of exposure to A. mellifera by sampling at multiple distances from apiaries, the lack of effect of distance on virome composition may have been a result of homogenous exposure to A. mellifera regardless of the distances sampled due to the presence of other apiaries in the area.

The factors that influenced the similarity between honey bee and bumblebee viromes were unclear, as the factors that significantly interacted with host genus and year, did not coincide with any meaningful convergence between the viromes of A. mellifera and B. impatiens. None of the viruses involved in these shifts were key honey bee pathogens, and many were minor members of the viral community (GO, CM, UP). Furthermore, it is not clear if poorly classified viruses such as these originated from honey bees, bumblebees, or other sources, such as other insects. The presence of various LSV related viruses is interesting, as one classification (lake sinai virus) did increase in abundance in B. impatiens in 2022, though no B. impatiens-specific cluster was observed (Fig. 4). Since the lake sinai viruses are a complex of diverse viral strains34 with phylogeny and host ranges that are still being resolved35, more systematic surveys would be needed to distinguish potential spillover with range expansion from natural host ranges in Bombus. Although surveys of DWV have led to the proposition that DWV has a broad host range due to spillover36 there remains little evidence to distinguish host relationships from incidental ones. Moreover, the lack of high titre, low diversity DWV in other insects29 suggest that these observations result from association and not spillover with host range expansion. Similarly, more expansive genomic studies would also be needed for many of the poorly characterised viruses (such as those dominating the B. impatiens RNA biased virome, Figs. 2 and 3) to determine the natural host ranges of these viruses.

Evidently, wild bumblebees are exposed to a wide range of honey bee viruses, but insights from other systems indicate that only a small minority of these viruses and exposure events could lead to spillover1,2,3,6,7 and therefore pose a significant threat to wild pollinators. It is likely that many cases of viral exposure between insects have occurred over long evolutionary timescales without spillover with range expansion, due to biological barriers that maintain distinct viral host ranges, and these dynamics must be distinguished from actual spillover threats. Our meta-transcriptome survey demonstrates an approach similar to that used in other spillover systems9 by which spillover with range expansion can be distinguished by the background noise of widespread viral exposure and dead-end spillover. Even in the case of DWV, where the virus can replicate in a new host that it frequently encounters in the wild, no spillover with range expansion may ever occur if the virus cannot surmount the biological and ecological barriers to it establishing a new maintenance host. There is a dynamic and complex virus system in wild insect pollinators that has been little explored, and a need to integrate genomic information into current efforts to study insect virus spillover in its entirety. Further genomic studies may yet find evidence of viral spillover with range expansion if they identify honey bee virus adapted variants that are specific to bumblebees and other pollinators.

Although this study shows a lack of spillover with range expansion between A. mellifera and B. impatiens, protection of bumblebees of conservation concern from potential pathogen spillover from honey bees is still a valid conservation action, particularly when protecting species at risk of extinction. It is important to note that dead-end spillover could still pose a threat to bee populations. Such a scenario may be possible, for example, in which exposure of bumblebees to honey bee viruses may lead to repeated, but self-limited, spillover without genetic adaptation to the dead-end spillover host. Low genetic diversity, which has been demonstrated in bumblebee species of conservation concern, could increase vulnerability to pathogen spillover more generally37,38,39. Due to the lethal nature of sampling for this study, bees of conservation concern were not examined. While this examination of B. impatiens is an important step in understanding the interactions between the RNA viromes of honey bees and bumblebees, B. impatiens appears to be expanding in range and is not expected to be representative of all bumblebee species, particularly those of conservation concern40,41. This study adds to a small but expanding view of insect RNA and to a limited extent DNA viromes and is an attempt to distinguish the largely overlooked outcomes of cross-species viral exposure between insects. Unsurprisingly, the common eastern bumblebee has a distinct viral community from honey bees, and the question must be raised as to what role virus-virus interactions play in these dynamics. With around 250 species of bumblebee globally, there is also much to learn of differences in viral communities among bumblebee species, particularly those of conservation concern. Establishing what the common state and composition of insect pollinator viromes (RNA and DNA) is essential, as it provides a baseline with which to compare proposed cases of spillover.

Materials and methods

Field sampling

Between the years 2021 and 2023 in Twin Cities, Minnesota USA, A. mellifera and B. impatiens were collected from three apiary sites between the months May and November.

At each site (Golf, Crosby, Vets) A. mellifera was sampled from colonies, repeated in replicates of 8 colonies per site every 3 weeks, collecting at least 30 worker bees per replicate. Bees, both A. mellifera and B. impatiens, were also collected from flowers during 3 collection periods (early July, early August, and late August) when B. impatiens colonies are at peak colony size. Bees were collected within 100 m of the apiary site, between 500 and 1000 m of the apiary site, and between 1500 and 2000 m of the apiary site. These distances were chosen to form a gradation of flower sharing, with fewer honey bees from the apiary site being present at farther distances, though there was no control for other potential apiary sites in the area that were not part of this study. When possible, four bees were collected from each distance area for each collecting period. Outside of the apiary sites, bees were collected from areas with flowering plants in roadsides or public parks. Lack of bee forage and low bee abundance in some areas resulted in some collection periods with fewer than 4 individuals per species and distance category. In 2021, bees were collected from all flowers. In 2022 and 2023, floral visitation was assessed before each collection period by observing bee visitation to flowers in the area for 20 min. Bee collection focused on flowers on which both honey bees and bumblebees were observed foraging. During times of low bee abundance, bumblebees were collected from flowers without observation of shared foraging by honeybees. Replicate samples were collected by year, collection period, distance from apiary, and bee species.

In 2023, two B. impatiens colonies were also sampled before and after transplantation near the Golf apiary site. One B. impatiens colony was initiated from a field-caught queen in the spring using standard rearing techniques42. The other colony was collected from the field and kept inside for 1 week then transplanted to site Golf. Colonies were sampled before transplantation (date 0), then repeatedly at multiple dates with one week between samplings (between July and August) from date 0 to 5–6 dates after transplantation, collecting 4 bees per sampling on average.

Dissociation and RNA extraction

Honey bees and bumblebees without ectoparasite mites (such as Varroa) and damaged wing or body parts were dissociated using a gentleMACS dissociator in 10 mL of molecular grade water using setting RNA 02.01. The resulting homogenate was centrifuged at 21,100 g for ten minutes at room temperature. Next, 150 µL of the supernatant fluid was retained for bead-based RNA extraction using a NucleoMag Virus RNA/DNA Extraction kit (Takara, Shiga, Japan), according to the manufacturer’s instructions and using a KingFisher Flex automated extraction machine (ThermoFisher, Walton, MA). The elution volume was 50 μL of molecular grade water.

cDNA synthesis

The template used for cDNA synthesis was 4 μL of extracted RNA per sample, which was annealed with 1 μL of 10 μM custom N6 template switching oligonucleotide (TSO; GCAGTGGTATCAACGCAGAGTACNNNNNN) and 1 μL of 10 mM dNTPs. The annealing mix was incubated at 70 °C for 5 min. Next the total volume of the annealed mix (total volume 6 μL per sample) was used as the template for first strand synthesis, to a total volume of 10 μL per sample. For each sample, 2.5 μL of template switching buffer (New England Biolabs, Ipswich, US), 1 μL of template switching enzyme (New England Biolabs, Ipswich, US), and 0.5 μL of a 75 μM TSO (AAGCAGTGGTATCAACGCAGAGTACrGrGrG) was used. This first strand synthesis mix was then cycled for 90 minutes at 42 °C, and finally for 5 min at 85 °C. Finally, 25 μL of PrimeSTAR GXL Premix (Takara, Shiga, Japan), two μL of 10 μM TSO primer (AAGCAGTGGTATCAACGCAGAGTAC), and 13 µL of molecular grade water were added to each sample to a total volume of 50 μL. The second strand synthesis was incubated at 94 °C for one minute, then for 30 cycles at 98 °C for one minute, 30 cycles at 60 °C for 15 se, and 30 cycles at 68 °C for 2.5 min.

Clean up of the cDNA was performed by adding 50 µL of CleanNGS DNA & RNA Clean-Up Magnetic Beads to 50 µL of the PCR-amplified cDNA, then incubated at room temperature for 10 min. After incubation, samples were centrifuged at 21,000 g force for ten seconds and then placed on a magnetic tube stand. The clear portion of the liquid was removed and discarded after the beads had collected on the side of the tubes, and 200 µL of 80% ethanol was added, left for 30 s, and then removed, and then this ethanol was repeated once more. The tubes were centrifuged for 10 s at 1000 rpm or until excess ethanol collected at the bottom and was then removed. The tubes were then left open to dry at room temperature for 1–3 min, then resuspended in 20 μL of molecular grade water, incubated for 10 min at room temperature before placing back on the magnetic stand and retaining the cDNA eluate.

Sequencing

Sequencing was performed using the Oxford Nanopore Technologies (ONT) sequencing technology, with an input of 12 uL of cDNA per sample, R10 flow cells and SQK-NBD114-96 kit according to the manufacturers’ instructions (using the kit SQK-RBK110-96 and its option of pooling the barcoded samples together). Only a single set of pooled barcoded libraries were used per flow cell. The sequencing run durations were 24 h and used the MinKNOW (v21.02.5) software and ran on a GridION (Oxford Nanopore Technologies, Oxford, UK). Basecalling was performed using Guppy (v4.2.0) with the high-accuracy model and with all default settings applied. Sequencing reads were filtered to a minimum length (≥200 bp) and Q-score (≥9) and demultiplexed by MinKNOW v24.02.16.

Contig assembly

Adapter sequences were removed using PoreChop v0.2.4 was used to remove the nanopore barcode adapter sequences. To generate metagenome-assembled contigs, the quality filtered reads were de novo assembled by Canu (v2.2)43(with parameters maxInputCoverage = 10,000 corOutCoverage=all corMinCoverage=0 minReadLength=200 minOverlapLength = 25 corMhapSensitivity=high genomeSize=5000). All contigs below 1000 base pair length were excluded from all analyses. To compare the lengths of viral contigs between A. mellifera and B. impatiens, contig lengths were visualised (Supplementary Fig. 4) using ggbeeswarm (v0.7.2).

Viral sequence identification

To identify and bin the viral contigs, the outputs of several tools were combined for assessment. To taxonomically classify the contigs, Kaiju (v1.9.2) was used (default parameters against both the Reference Viral Database44(RVDB; v27) and NCBI non-redundant (nr) datasets built into Kaiju). To assist with contig binning, MMseqs2 (v15.6f452; easy-cluster --min-seq-id 0.9 -c 0.8 --cov-mode 5) was used to assign viral contigs to clusters. All contigs were queried against the Virus Orthologous Groups Database (VOGDB, release 217) and RVDB using diamond BLASTx (v2.1.8.162; --range-culling --max-target-seqs 1 -F 15 --evalue 1e-20 --sensitive). For all contigs, checkV (v1.0.1; end_to_end) was run, and the genome quality assessment was also used to filter and bin the contigs. The combined output of these tools was used to assess the contigs and place them into bins. Binning the contigs down to species level (or as low a taxonomic level as possible) was performed according to the Kaiju classifications, and the final bins were named after these classifications. These classifications represent the nearest best hit in the databases, and do not represent accurate classification at lower taxonomic levels for poorly represented viral classifications. The contig binning assessment was performed as follows: contigs were binned as Non-viral if they had the CheckV warning “no viral genes”, and no viral hit against Kaiju, and no BLASTx hit against RVDB, and no BLASTx hit against VOGDB. Otherwise, contigs were considered viral. Some contigs were removed from the viral bin if they had a combination of CheckV “no viral genes” or CheckV “Low-quality” and no BLASTx hits versus VOGDB and RVDB. Furthermore, contigs with certain VOG hits associated with cellular organisms, such as gag pol retrotransposons were removed from viral bins. Due to taxonomic discrepancies between NCBI and the International Committee on Taxonomy of Viruses (ICTV), various assignments by Kaiju were manually corrected to align with those of ICTV.

Phylogeny

Gene predictions were generated using Prodigal (v2.6.3, parameters: -g 11 -p meta), followed by Pfam annotation of genes using Interproscan (v5.23-62.0, protein query). Protein sequences were grouped by bin and Pfam, and sequences corresponding to RNA dependent RNA polymerase (RdRp) were aligned using MAFFT (v7.525) (parameters: --auto). Before alignment, proteins less than half the average length of the viral cluster were removed. Alignments were trimmed using TrimAl (v1.4.rev15, parameters: -automated1) and phylogeny performed using IQTREE (v2.2.6, parameters: -m MFP) with 1000 bootstraps. Trees were visualised using ggtree (v3.6.2) in R for iflaviruses (Fig. 4B) and other taxa (Supplementary Fig. 7: apis rhabdovirus, Dicistroviridae, Iflavirus aladeformis, bactrocera tryoni iflaviruses, Iflavirus sacbroodi, nege-like virus; Supplementary Fig. 8: partiti-like virus, unclassified phasmaviridae, picorna-like, Sinaivirus, virga-like).

Metagenomic analyses of the RNA virome

Prior to mapping, all assembled contigs were de-replicated, independently for each sample, using CD-HIT (v4.8.1, parameters: -c 0.9 -s 0.8). The barcode-trimmed raw reads were mapped using Minimap2 (v2.26, parameters: -ax map-ont --secondary=no), mapping the reads of each sample independently against all concatenated contigs from all samples. Secondary read mappings were removed from the mapping results using SAMtools (v1.18; -h -F 2308). All contigs (not just viral ones) were included in the mapping, but non-viral contigs were filtered out of the downstream analyses. To normalise the read counts by library size, the R package metagenomeSeq (v1.40.0) was used to perform cumulative sum scaling (CSS), without transformation of the CSS counts. To examine the yield of reads for viral taxa, the CSS counts of barcode trimmed raw reads mapped against the classified contigs were visualised as a heatmap Supplementary Fig. 5). The CSS counts of mapped barcode trimmed raw reads were converted to a phyloseq object in R (v4.2.2) using the package phyloseq (v1.42.0), with the incorporation of the Kaiju classifications as taxa, and the sample metadata. Using this phyloseq object as a basis, the R package microViz (v0.11.0) was used to perform principal component analysis (PCA), redundancy analysis (RDA), correlation heatmaps, and composition plots. All plots were visualised using R’s ggplot2 (v3.5.0) or base R. The PCA and RDA plots were generated using different variables to identify which viral taxa or experimental variables were contributing to the composition of viral communities. They were used to identify how viral community similarity was affected by viral taxa across all samples (Fig. 2 - shown split by collection year in Supplementary Fig. 3), for the transplanted B. impatiens colonies (Supplementary Fig. 6A), and for the apiary, distance from apiary, and flower genus from which the samples were collected (Supplementary Fig. 1). To further examine the association of specific viral taxa with host genus and collection year, correlation heatmaps were also generated, with the Pearson correlation coefficients indicating the strength of association with either bee species (Supplementary Fig. 2). The composition of viral communities was visualised as bar plots for all samples (Fig. 3) and the transplanted B. impatiens colonies (Supplementary Fig. 6B).

Network analyses

To generate percent similarity scores, fastANI (v1.33, parameters: --fragLen 200) was run on the whole contigs, all versus all. For RdRp, the protein sequences were queried against themselves using diamond BLASTp. The similarity scores were imported into Cytoscape (v3.10.2) and visualised with the organic layout, and hiding all edges below 98% identity. Networks were generated for this study’s sequences with some reference sequences for average nucleotide identity (Fig. 4A) and protein percent identity (Supplementary Fig. 9), and also for reference virus sequences downloaded from NCBI, for nucleotide sequences over 200 bp length and recovered from Bombus spp., Apis mellifera, and Apis cerana (Supplementary Fig. 10) and RdRp protein sequences predicted from these sequences (Supplementary Fig. 11). Nucleotide and RdRp networks were also generated for reference sequences over 4000 bp length and derived from various insect taxa (Fig. 5). Reference sequences were accessed in June 2024 from NCBI Virus, selecting nucleotide sequences under the taxids for Iflavirus sacbroodi: 3048309, 89463, 886669, Iflavirus aladeformis: 198112, 3047792, 2951794, 2498590, and Triatovirus nigereginacellulae: 3047684.

Statistical analysis

Permanova was performed using adonis2 (method = “bray”) in R, using an input matrix of the CSS read counts for each contig classified by viral taxa. For permanova, interactions between the variables host genus, collection year, distance from apiary, apiary, and flower genus were included (Supplementary Table 1). These variables were included to assess whether viral community composition varied across the years, between sites (apiaries) between A. mellifera and B. impatiens. Distance was considered because the viral community was expected to be impacted by decreasing A. mellifera presence with increased distance from the apiaries. Flower genus was expected to alter viral community composition due to differences in the physical properties of the flowers and the levels of A. mellifera and B. impatiens visiting different flower types. In addition, other flower properties were included in the permanova, including flower depth (shallow = < 3 mm, medium = 4–6 mm, deep = >8 mm), flower shape (classified as composite, cup, pea, tube), and the ratio of A. mellifera and B. impatiens observed visiting that flower species during the study. Contig lengths were tested for differences in length using a Wilcox test (wilcox.test) in base R (Supplementary Table 2).

Rarefaction analysis

Using rarecurve in the R package Vegan, sequence sampling curves were plotted against virus species accumulation (Supplementary Fig. 12), which indicate that in all samples viral diversity sampling flattened out during sequencing.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.