Small RNA genomics of Aedes aegypti mosquitoes discovers infectious viruses that trigger an RNA interference response

Abstract

We report a global survey of viral small RNAs (vsmRNAs) from >200 Aedes aegypti samples to identify many mosquito viruses that actively infect this prominent arboviral vector. Ae. aegypti viruses in the Americas are abundant, with some displaying geographical boundaries. Viruses infecting Asian Ae. aegypti are similar to those in the Americas and reveal the first wild example of dengue vsmRNAs. African Ae. aegypti display vsmRNAs from viruses unique to these African strains. Academic lab colonies generally lack viruses, yet two commercial strains are deeply infected by a tombus-like virus that is related to plant viruses. Comparing matched viral long RNAs to vsmRNAs reveal viral transcripts evading the mosquito RNA interference (RNAi) pathway. By infecting mosquito cells with Ae. aegypti homogenates, we generate stably infected cell lines which produce vsmRNAs that were comparable to native mosquito vsmRNA patterns. Lastly, we demonstrate that these stably infected mosquito cells producing vsmRNAs can exert gene silencing of reporters bearing viral sequence segments, providing a potential explanation for how Ae. aegypti can resist viral infections. This vsmRNA genomics approach in Ae. aegypti can add to existing vector surveillance approaches by discovering new viruses that persist in mosquito populations.

This study is a small RNA genomics survey from global Aedes aegypti mosquito samples, and detected replicating viruses. This study then isolated several mosquito viruses into cell cultures to show that viral small RNA are active in gene silencing.

Introduction

Aedes aegypti, the yellow fever mosquito, has a global reach across many major human locales. This insect is a health threat as a prominent vector for many arthropod-borne viral (arboviral) diseases like dengue fever. Municipal vector control organizations conduct routine molecular surveillance of arboviruses from trapped Ae. aegypti mosquitoes but are only able to assay known viruses with conventional RT-PCR methodologies¹. Recent research efforts applying high-throughput RNA sequencing have now led to a large expansion in the mosquito virome lists^2–6. However, open questions remain as to how persistent are insect viruses within mosquito colonies, how insect viruses interact with mosquito immune responses and how frequently can insect viruses be vertically transmitted.

During virus replication, viral double-stranded RNAs are processed into viral small interfering RNAs (siRNAs, ~18–23 nucleotides long) that are loaded into Argonaute (AGO) proteins⁷. A second RNAi pathway that mainly regulates Transposable Elements (TEs) involves the Piwi proteins that bind Piwi-interacting RNAs (piRNAs, ~24–35nt long), many of which have sequences complementary to TEs⁸. Since many TEs are evolutionarily related to viruses, mosquito Piwi proteins may generate viral piRNAs similar to how TE piRNAs are made. The viral siRNAs that are antisense to the viral mRNA will trigger AGO cleavage of viral mRNAs. Viral piRNAs may have antiviral properties in mosquitoes^9,10, but this response may be complicated by possible virus-to-virus interactions¹¹, and the expression of small RNAs deriving from Endogenous Viral Elements (EVEs)^12–19, the latter of which may feed into the RNA interference (RNAi) pathways in mosquitoes to enable virus resistance.

Vertical transmission and infection rates of medically-relevant arboviruses in Ae. aegypti have been examined for dengue and Zika viruses, among others²⁰. Even when infection rates are high, vertical transmission rates can be low if virus infection does not extend to the gonads^21–24. Vertical transmission of arboviruses like Zika virus in laboratory Ae. aegypti infections is possible, but at low rates^25,26. In addition, tracking the prevalence of virus infections in mosquitoes in the wild is a complex epidemiological challenge due to fluctuating levels of virus infection rates, even during an outbreak^27–29.

The ability of a virus to persist via vertical transmission in mosquito populations raises the concern of a mosquito-borne virus transmitted to humans^30,31. Yet, tracking persistent viruses in wild and small pools of mosquitoes by qualitative RT-PCR during regular surveillance can be limited by this technique’s inability to distinguish between high and low viral loads. We hypothesize that small RNA metagenomics can add dimensions to regular surveillance approaches by discovering the extent of viruses infecting a mosquito sample and provide a representation of vertical transmission by persistent virus infection in mosquito populations.

A recent study by Olmo et al. 2023 profiled Ae. aegypti and Ae. albopictus small RNAs from South America, Asia, and Africa-derived strains¹¹. However, the study’s viral small RNAs (vsmRNA) detection was not optimized and missed some insect viruses that our Mosquito Small RNA Genomics (MSRG) resource could detect as widespread insect virus persistence in mosquito cell cultures and in laboratory mosquitoes³². Thus, the Olmo et al. 2023 study primarily focused on testing whether dengue virus (DENV) competence could be influenced by two known mosquito insect viruses, Humaita-Tubiacanga virus (HTV) and the bunyavirus Phasi-Charoen-Like virus (PCLV)¹¹. The open question remains: how prevalent are other viruses in Ae. aegypti mosquitoes around the world that are capable of triggering a small RNA response?

To answer this question, we conducted an improved survey of Ae. aegypti mosquito small RNAs, comprising new laboratory strains and wild-caught and lab colonies from the Americas, Asia, and Africa. Although precise transmission dynamics maintaining viral diversity in lab strains are incompletely understood, and laboratory colonization may impact virome structure of surveyed mosquitoes, the early generation lab colonies are still likely to reflect and are valuable in expanding the known diversity of mosquito-associated viruses. Furthermore, detection of high levels of vsmRNAs may signal mosquito virus persistence or high enough virus levels in the animal to induce a physiological RNAi response.

Our results significantly expand the catalog of insect viruses that are capable of propagating in mosquitoes and generating a small RNA response. We demonstrate experimental evidence for the transmissibility of these novel insect viruses to mosquito cell cultures, where the generated vsmRNAs produce an RNAi response that is capable of gene silencing. We propose that detection of vsmRNAs in mosquitoes persisting across generations represents a sufficiently-high load of viral infection, demonstrating the value of a small RNA genomics approach in adding to current molecular surveillance efforts to detect potential emerging pathogens vectored by mosquitoes.

Results

Because the global COVID-19 pandemic has reshaped and limited field-work efforts^33–36, we undertook the most feasible option in the post-pandemic era of receiving Ae. aegypti mosquito samples from collaborators shipping to our Boston University lab for total RNA extraction and small RNA library preparation. Some samples were field collections from vector control districts, while other samples were from lab colonies established after field collection, with all generation numbers and lab origins noted in Supplementary Data 1. Although mosquito lab colonies after extended generation times may have altered physiological characteristics³⁷, extensive literature supports mosquito lab colonies as useful proxies of the founding mosquitoes origins^38–48. In our study, any unique viruses we could still detect in these limited colony generations would be reasonable reflections of a persistent, vertically transmitted infection.

Most mosquito RNAs were analyzed from whole mosquitoes to optimize for throughput. Published small RNA libraries from Olmo et al.¹¹ and our previous study³² were also integrated into our analysis of vsmRNA profiles. In total, we compiled >280 small and long RNA sequencing datasets of Ae. aegypti sampled from locations originating in North, Central and South America, Asia, and Africa, as well as from a select set of laboratory strains (Fig. 1). All datasets were analyzed by our MSRG pipeline and primarily focused on Insect Specific Viruses (ISVs) and arboviruses, while TEs and piRNA cluster loci analyses will be presented in a future study.

Fig. 1. A mosquito small RNA sequencing approach to discover mosquito-host RNA-interference (RNAi) responses to insect viruses.

a Overview of the Mosquito Small RNA Genomics (MSRG) pipeline applied to a survey of whole mosquitoes from Americas, Asia, Africa and laboratory strains. b Implementation of the VirusDetect program with updated GBVRL and custom databases for comprehensive mosquito virus detection. c Summary tabulation of the samples and RNA libraries analyzed in this study.

To address library quality control (QC), we first only selected libraries for analysis with at least 10 million (M) reads before adaptor trimming. There was an average depth of ~40 M reads across all libraries (see Supplementary Data 1 and Supplementary Data 2 for library sequencing depth statistics). We also tracked an EVE in Ae. aegypti named AEFE1¹⁷, whose small RNAs are expressed in all mosquito samples we analyzed and are represented by primarily antisense piRNAs seen as blue peaks in coverage plots (Supplementary Fig. S1). Then, we inspected the read length distribution profiles for each library to track the expected peaks for miRNAs and piRNAs (Supplementary Fig. S2). Presence of AEFE1 and an expected read length distribution profile verified that libraries were of good quality. Lastly, viruses with less than 10 small RNA reads per million (RPM) were filtered out. Main figure bubble plots show all small RNAs (18–35nt), whereas breakdowns of the siRNAs (18–23nt) and piRNAs (24–35nt) are shown in Supplementary Fig. S3.

Augmenting the MSRG pipeline with improved virus discovery

Mosquitoes in the culicine clade exhibit notable somatic piRNAs that reflect the broad diversity of small RNA sequences in whole animal total RNA⁴⁹. This breadth means that small RNA libraries cannot be queried efficiently or specifically against databases such as the GenBank Virus Reference List (GBVRL) that have become flooded with recent massive virome surveys^4,5 and coronavirus variant genomes⁵⁰. In GBVRL version-249, there were ~7.4 million records, and 53.7% of these were beta-coronaviruses. To overcome this inefficiency and provide scalability, MSRG utilizes curated lists of viruses for small RNAs to be mapped against using BowTie (v1)⁵¹.

To address the limitations of the manually curated virus database in the MSRG, we devised an auxiliary pipeline to first analyze all Ae. aegypti small RNA libraries with the VirusDetect program⁵² (Fig. 1b). The VirusDetect program first performs de novo assembly of small RNAs into long contigs that can then query GBVRL efficiently and specifically to yield the count and coverage of the contigs against the most updated virus list in GBVRL (Supplementary Fig. S1b, S1c). In this study, we updated the GBVRL in VirusDetect and locked it at version-249 from June 2022. The VirusDetect results confirmed the presence of viruses already present in the 2019 MSRG virus database, while also revealing new viruses that we then added to our MSRG virus database list (Supplementary Data 3).

We then set up additional VirusDetect queries of our mosquito small RNA datasets against the Third-Party Assembly (TPA) and Metagenomics Assemble Genomes (MAG) databases that are distinct from GBVRL. This step was added because when VirusDetect was run on GBVRL alone, a false-positive call was made for a tombus-like virus isolated from a Hyposignathus monstrosus bat⁵³ due to the low contig coverage (Supplementary Fig. S1d, S1e). The additional TPA and MAG runs discovered the Tiger Mosquito Bi-segmented Tombus-Like virus⁵⁴ (TMBTLV) as the true source of these small RNAs, as well as revealing a new Aedes partiti-like virus⁵⁴. This step provided additional confidence in the identity of viruses. Phased and ping-pong piRNA biogenesis signatures⁵⁵ were evident in several of the viral smRNA profiles (Supplementary Fig. S1f). In total, the VirusDetect analysis added 94 new viruses to our MSRG virus list that we locked in October 2023 for the completion of this study (Supplementary Data 3, Source Data Zip File).

Widespread mosquito viruses can span continents while also displaying geographic boundaries

With the help of municipal vector control departments in California and Miami, Florida, along with field collections and lab-maintained colonies from the DeGennaro, Olson, Lambrechts, Manning, Calvo, Chen, and Gloria-Soria labs, we assembled a diverse collection of Ae. aegypti samples from the American and Asian continents to generate and sequence small RNAs from whole mosquitoes. We also integrated publicly available Ae. aegypti small RNA libraries from Suriname, Brazil, and Singapore¹¹.

The most widespread viruses across Ae. aegypti small RNA samples of the Americas were PCLV and HTV (as noted previously in Olmo, et al.¹¹), along with the Ae. aegypti Anphevirus (ANPHV) that was first described in a Florida strain⁴⁹. Remarkably, PCLV and HTV were most frequently found in Ae. aegypti of the southern fraction of the Americas, while ANPHV displayed a bias for the northern fraction of the Americas (Fig. 2a, b). Florida, Mexico, and Caribbean locales reflected a ‘mixing zone’ for these three viruses. Verdadero virus in the Americas was found exclusively in this mixing zone with PCLV, HTV, and ANPHV (Fig. 2b).

Fig. 2. Ae. aegypti in the Americas display diverse insect vsmRNA responses with geographic boundaries, whereas Asia Ae. aegypti share these vsmRNA patterns and reveal dengue vsmRNAs from a wild isolate.

a Map of the Americas and Asia where the Ae. aegypti samples have originated from, with geographic boundaries delineated by the distribution of ANPHV (purple shading in the Americas map) and HTV/PCLV (green shading in the Americas map) reflecting what is displayed in the bubble plot. Map data ©2025 Google. b Bubble plot of vsmRNAs from Americas and Asia Ae. aegypti samples. Number reads per million is reflected by bubble diameter, and color represents strand bias of reads, red is plus strand biased, blue is minus strand biased. c Example coverage plots of PCLV, HTV, and ANPHV vsmRNAs organized by zone. d DENV vsmRNA coverage from a wild Singapore isolate compared to other DENV vsmRNA patterns in Ae. aegypti from a lab-injected infection of DENV2 by the Myles lab and blood-fed infections of DENV4 by the Marques lab.

VsmRNA patterns and abundances fluctuated widely between individual mosquito isolates (Fig. 2c, Supplementary Fig. S4), but ANPHV and HTV small RNAs had a clear plus strand bias, especially for the viral structural genes. In contrast, PCLV small RNAs displayed a stronger antisense bias and mostly originated from the ‘Small’ and ‘Medium’ virus segments. Even though many of the samples in these regions contained the same viruses, the coverage pattern of these viruses displayed remarkable differences. There were variations in the patterns of Verdadero vsmRNAs between Mexico-originating mosquitoes and several different mosquito samples from Miami, Florida (Supplementary Fig. S5). The Binegev-like virus and Renna virus were found from the west to the east coasts of the United States, and they also displayed interesting variations in vsmRNA patterns between samples (Fig. 2b, Supplementary Fig. 5b).

Despite some of these viruses displaying geographic boundaries within the American continents, many were also found in common with the Asian samples, such as HTV, PCLV, and ANPHV (Fig. 2b). For most of these shared viruses, vsmRNA patterns in the Asian samples resembled that of their American counterparts (Fig. 2c, Supplementary Fig. S5d, S5e). For other viruses, such as the Partiti-like virus-1 Jane strain that we detected in the ThaiKP sample, coverage patterns varied across samples. The ThaiKP Partiti-like virus-1 vsmRNAs were strongly biased for the plus strand of the RdRP gene in contrast to both plus and minus strand siRNAs against this virus in a Singaporean wild isolate (Supplementary Fig. S5f). A Florida sample from the Jiggins lab had Partiti-like virus-1 Jane that looked more similar to the Singapore sample, while samples from New Orleans, Louisiana had both siRNAs and piRNAs that were only slightly plus strand biased. The molecular implications of these and similar diverse fluctuations in vsmRNAs from HTV (Fig. 2c) remain unknown.

Dengue vsmRNAs identified in a wild Ae. aegypti Asian isolate

Miami, Florida, and various places in Brazil are sites of recent dengue fever outbreaks⁵⁶, yet no DENV small RNAs were detected in any of these samples, including those from Ae. aegypti sampled by the Florida Department of Health (FDOH) in the vicinity of known dengue fever patients. The lack of DENV small RNAs mirrors the sporadic and challenging detection efforts for DENV in surveillance of mosquitoes to precede a human disease outbreak^1,27–29. However, all these samples revealed that multiple insect viruses can simultaneously infect and generate a robust RNAi response of vsmRNAs in a small mosquito cohort (Fig. 2b).

Surprisingly, our MSRG pipeline detected the first case of dengue vsmRNAs from a wild mosquito isolate – a Singaporean sample whose RNA was sequenced in Olmo, et al.¹¹ (Fig. 2d). DENV1 siRNAs in this Singaporean Ae. aegypti were more abundant than viral piRNAs, reflecting the same trend as DENV infections in mosquito Aag2 cells³². These natural DENV1 small RNAs also accumulated to the same level as DENV2 and DENV4 infections in lab Ae. aegypti experiments^11,57, reflecting the likelihood of such a high DENV1 load in this Singaporean sample that enables our MSRG pipeline to detect so many dengue vsmRNAs.

African Ae. aegypti colonies harbor a persistent African-specific mosquito virus

Most Ae. aegypti samples in this study thus far belong to the Ae. aegypti aegypti subspecies that have higher human biting proclivity and better Zika virus (ZIKV) vector competence compared to the contemporary African subspecies Ae. aegypti formosus^40,42. Most African colonies in this study are subspecies formosus, but some are subspecies aegypti (THI, NGO, CVerd), while others are substantially admixed (KUM, OGD). Distinct genetic backgrounds between aegypti and formosus subspecies are one likely explanation behind the differences in these subspecies⁴⁰, but the persistent viruses generating vsmRNAs in formosus subspecies were previously unknown.

Our small RNA profiles revealed that PCLV’s global reach extends into several African colony strains primarily on the western African coast, which may be biased to the limited access to colonies only from these African locales (Fig. 3a, b). The PCLV small RNA coverage patterns were also more biased towards the ‘S’ and ‘M’ viral segments with plentiful antisense viral piRNAs, while the ‘L’ segment mainly generated sense viral piRNAs (Fig. 3c). Unlike Asian and Central and Southern American aegypti strains, where HTV frequently accompanied PCLV in causing a small RNA response in the same mosquito sample, formosus strains only had a few HTV small RNA signatures, all of which were found exclusive of PCLV (Fig. 3a).

Fig. 3. African Ae. aegypti colony strains carry vsmRNAs from Formosus virus, which is unique to the African continent.

a Bubble plot of vsmRNAs from Africa Ae. aegypti colony strains. Number of reads per million is reflected by bubble diameter, and color represents strand bias of reads, red is plus strand biased, blue is minus strand biased. Dashed pink line boxes mark the PCLV and FORMV noted in panels (c) and (d), respectively. b Map of Africa locations where the Ae. aegypti colonies or samples originated. Map data ©2025 Google. c Coverage plots of PCLV small RNAs from a selection of African Ae. aegypti showing high M-fragment piRNAs rivaling the S-fragment piRNAs. d The FORMV vsmRNA coverage from African Ae. aegypti colonies from the McBride lab and an independent Kedougou, Senegal sample from Olmo et al.²³. The black arrow points to male-specific viral piRNA species. e Three examples of FORMV long RNAs sequenced from matched samples in (d).

Formosus virus (FORMV) displayed the most striking persistent small RNA response in 12 samples from native-range colonies (Fig. 3a). This rhabdovirus has a ~ 12.2 kb genome that was initially absent from GBVRL database. It was deposited in the TPA (accession BK059424) by Parry, et al.⁵⁴, from a metagenomics assembly of transcriptomes from a different African lab colony not in this study originating from Bundibugyo, Uganda⁵⁸. Importantly, four additional African Ae. aegypti samples from Olmo et al.¹¹, that were established and maintained separately from the main set of native-range colonies, also displayed FORMV small RNAs with similar coverage patterns (Fig. 3d, i.e., KED_02_JM and KED_Female_LM). Since most of the native-range colonies here were sampled after at least seven generations in the lab, we conclude that FORMV is persisting and being transmitted vertically within the colonies.

FORMV primarily generates viral piRNAs from the sense strands of the NP, HP, and GP genes in the 5’ half of the viral genome, while the RdRP gene seemingly evades piRNA production (Fig. 3d). Conversely, long RNA sequencing indicated the plus-strand RdRP gene is transcribed as highly as NP, HP, and GP genes (Fig. 3e). We also detected robust minus strand long RNAs indicative of replication of the negative-strand rhabdovirus genome, yet much fewer viral piRNAs were generated from this minus strand (Fig. 3d). Future studies will investigate how the RdRP gene transcript and negative-strand genomic transcript evade the RNAi machinery. Lastly, we observed some sex-specific Formosus viral piRNA accumulation patterns in males at the 5’ end of the NP gene, even though long viral RNA patterns look similar between the sexes.

Academic and commercial laboratory strains of Ae. aegypti

Limited insect viruses were found in academic lab strains from these new Ae. aegypti small RNA libraries as well as from previously examined lab strain datasets (Fig. 4a). Our first MSRG pipeline study³² also revealed very few insect viruses in these strains, but the expanded virus list and low viral reads in this study raise our confidence in the sparseness of viruses in these strains.

Fig. 4. A vertically-transmitted tombus-like virus generates abundant vsmRNAs in commercial Ae. aegypti lab strains.

a Bubble plot of vsmRNAs from lab strains. Reads per million represented by bubble diameter, strand bias represented by color, red is plus strand biased, blue is minus strand biased. See Supplementary Data 1 for sample details. Lab initials: BZL=Benzon Research, MY = M. Younger, DB = D. Brackney, JM = J. Marques, ZT = Z. Tu, GH = G. Hughes, TC = T. Colpitts, BH = B. Hay, GP = G. Pijlman labs. b Coverage plots of long RNAs compared to small RNAs for viruses from the BZL strain of Ae. aegypti females and dormant eggs. c Coverage plots of TMBTLV small RNAs from GP lab strains also infected with Zika virus (ZIKV). d Scatterplot comparing matched small and long RNA libraries from Ae. aegypti. Sequencing RPM are plotted on a logarithmic scale. Sample dots are colored by sex, and clustered samples are in labeled ovals. e Coverage plots of two Florida exhibiting abundant long RNA signal for the Toti-like virus but negligible vsmRNAs in the upper plots that contrast both long and small RNAs against an R1-Ele4 TE. f Coverage plots of long RNAs versus small RNAs for the Formosus virus and R1-Ele4 TE from both males and females of the ENT African colony.

In contrast to these academic lab strains, two independent commercial lab strains from Benzon Research Labs (BZL) and Bayer AG (GP)⁵⁹ contained massive amounts of TMBTLV small RNAs, with viral siRNAs more abundant than viral piRNAs (Fig. 4b, c). Coverage patterns of the TMBTLV small RNAs differed greatly between the two commercial strains. The BZL strain vsmRNAs were biased for the plus strand and mainly derived from the RNA dependent RNA Polymerase (RdRP) gene, while the GP strain displayed a much greater accumulation of minus strand siRNAs across the entire TMBTLV genome and had piRNAs still biased for the plus strand. All GP strain samples were co-infected with ZIKV, but control non-Zika-infected GP strains small RNA data were not available, so ZIKV’s contribution to the massive TMBTLV siRNAs cannot be resolved until future experiments can be conducted.

Commercial labs often raise Ae. aegypti alongside many other arthropods for the testing of pesticides. Benzon Research Labs states that their BZL strain was derived from an USDA “Gainesville” strain from 1994. The continuous rearing in their facility for over 25 years may have contributed to this strain contracting additional viruses not seen in academic lab strains (Fig. 4b, Supplementary Fig. S6a, S6b). These viruses must persist by vertical transmission, because the same vsmRNAs in adult BZL females were also detected in BZL eggs (Fig. 4b), although the eggs had lower levels and more minus strand vsmRNAs. For some mosquito viruses, the small RNAs originated from all genes in the viral genomes, but for a partiti-like virus and TMBTLV, the RdRP genes served as the primary source of small RNAs.

Mosquito viruses related to plant viruses lack matches to Aedes EVE piRNAs

From our vsmRNA genomics survey, several mosquito viruses are related to plant viruses, such as tombus-like viruses and partitiviruses. Partitiviruses were originally discovered to infect plants, protozoans, and fungi, while tombusviruses are a group representing the tomato bushy stunt viruses^60,61. One partitivirus is the Verdadero virus found in the American strains that was first described infecting Ae. aegypti as well as Drosophila⁶². Verdadero virus small RNAs are primarily plus strand biased and cover the viral genome evenly, except for in Poza Rica samples which have more small RNAs coming from RNA2 (Supplementary Fig. S5b). A second partitivirus we detected is the Partiti-like virus-1 Jane strain⁵⁴ that interestingly has a wide variety of vsmRNA coverage patterns differing between American and Asian samples (Supplementary Fig. S5f) as well as a laboratory strain (Fig. 4c).

Two tombus-like viruses, TMBTLV from the two commercial lab strains and Liverpool Tombus-like virus (LTLV) from the American strains, were originally reported in the same metagenomics study⁵⁴. Like Verdadero virus, TMBTLV patterns differed greatly depending on the host samples (Figs. 4b, 4c). LTLV patterns in mosquitoes from Recife, Brazil suggest one route by which these plant-related viruses could be shifting from a plant to an insect host. Viral siRNAs and piRNAs for LTLV were much higher in the midguts compared to the total abdomens of an Ae. aegypti strain from Recife (Supplementary Fig. S5a). The abundance of LTLV in the digestive system suggests that mosquitoes feeding on plant sap nutrients could be a mechanism for the transfer of plant-related viruses to an insect host^63–65. When comparing the genomes of our TMBTLV and FORMV isolates to the initial genomes assembled by Parry et al.⁵⁴, we detected 25 total protein-coding mutations in the ~4.5 kb TMBTLV genome, but only 11 protein-coding mutations in the ~12.2 kb FORMV genome (Supplementary Fig. S7a–S7d).

Previous studies have suggested Ae. aegypti may resist virus infection with RNAi defense mechanisms^57,66–72, which may include EVE piRNAs with existing complementarity to viral genes that could provide an innate antiviral repression against virus replication^14–16. We applied a similar bioinformatics approach described in the Supplementary Text Discussion that defined EVEs generating significant piRNAs^12,14,15,18. Our analysis shows that all these EVEs were in an antisense orientation and generated antisense piRNAs against CFAV, ANPHV, GTTV, and FORMV (Supplementary Fig. S7e–S7g). Our MSRG pipeline uses a strict BowTie-v1.0 mapping of the small RNAs (i.e., it can clearly distinguish vsmRNA production for each DENV serotype despite an average 68.8% sequence similarities between the virus serotypes)³², and there are still many small sequence differences between the EVEs and the viruses. Our approach specifically discriminates vsmRNAs from EVE smRNAs and mitigates concerns of confusion because these results show that vsmRNAs primarily map in the sense-strand orientation just to the virus genomes while the EVE-derived smRNAs only map to the AeAegL5 genome EVEs that are orientated in the antisense orientation with many mismatches to the virus genomes.

Although we observed significant variation in small RNA expression levels from Ae. aegypti EVEs defined in the Crava et al. study¹² across our mosquito samples (Supplementary Data 4), this operating method was unable to detect EVEs against the mosquito viruses related to plant viruses such as TMBTLV, Verdadero, Partiti-like virus Jane isolate, nor LTLV. Although EVE discovery and characterization are not the primary focus of this study like in other studies^12–19, we cannot yet explain why these particular plant-related mosquito viruses lack apparent EVE piRNAs. Nevertheless, we are confident that the combination of our MSRG pipeline and study approach as detailed further in the Supplementary Text Discussion and subsequent results address the concern that virus discovery via small RNA genomics will not be confounded by known and unknown EVEs.

Insights into mosquito virus replication and RNAi dynamics from matched long RNA sequencing

Next, we asked if mosquito viruses like FORMV exhibit interplay between vsmRNAs and total RNAs indicative of a broader picture of dynamics between viral replication and vsmRNA biogenesis. We compared in a scatterplot the log-transformed total and small RNA levels between ~75 matched samples for each mosquito virus (Fig. 4d). We noted four interesting virus groups in this analysis. The first group is AEFE1 EVE measurements from adults all clustering together, which displayed more small RNAs than total RNAs, and reflects the care we took to generate and sequence these RNA libraries as reproducibly as possible.

The next two groups were viruses in samples that displayed significant viral total RNAs but few small RNAs, and vice versa significant vsmRNAs without much total RNAs. In several Florida isolates that were captured by municipal vector control surveillance, the Aedes Toti-like virus appeared to replicate and express viral genes effectively, perhaps before the mosquitoes could mount an RNAi response with small RNAs (Fig. 4d). Notably, these Florida mosquitoes had robust small RNA responses to endogenous TEs like R1-Ele4 as well as other viruses like Verdadero virus, validating that the lack of Aedes Toti-like virus small RNAs is not merely a technical error (Fig. 4e). More interestingly, there were some virus cases like FORMV in the Entebbe, Uganda colony (ENT), in which both males and females only generated vsmRNAs. This may be linked to the complete loss of viral long RNAs (Fig. 4f), despite the long RNA libraries still tracking the TE long RNAs.

The fourth standout group is viruses in the BZL strain eggs which display massive levels of long RNA reads from both plus and minus strands of the various viruses in this strain (Fig. 4b). For both TMBTLV and an Aedes Binegev-like virus, the long RNA reads were more than an order of magnitude greater than the small RNAs in both the eggs and whole adult females. Despite the production of abundant viral piRNAs and siRNAs in the adult whole female and maternal contribution of these vsmRNAs to the eggs, virus silencing does not appear effective, which could allow for efficient vertical virus transmission from adult female to egg.

Infectious capacity of metagenomically assembled virus entries from RNA sequencing

Although TMBTLV and FORMV were in the TPA and MAG databases within GenBank⁵⁴, they were not included in the GBVRL at the time of our initial analysis, raising the question of whether these entries represent truly infectious viruses. To bolster our RNA sequencing findings, we sought to isolate viruses from filtered mosquito homogenates to infect mosquito cell cultures (Fig. 5a). If the virus infections in cell cultures were deemed stable, we could then sequence the virus genomes for GBVRL submission and sequence vsmRNAs from infected cells to compare against the mosquito vsmRNAs.

Fig. 5. Transfer of viruses from mosquito lysates to infecting and replicating in mosquito cells.

a Our methodology to molecularly validate the small RNA detection of ISVs are true viruses that can be isolated and verified for triggering the RNAi response in mosquito cells. b RT-PCR detection of TMBTLV RNAs S1 and S2 during multiple rounds of blind passaging, starting with BZL mosquito homogenate as the initial virus infection source placed onto C6/36-NL and Aag2 mosquito cells. The ladder is the 1KbPlus DNA ladder, and uncropped gels are in the source data files. c Virus infection kinetics measured in C6/36-NL and Aag2 cells over 12 days using droplet digital PCR (ddPCR). Flasks with 1 million cells were infected on Day 0 with 20 K viral copies per infection. Virus stocks are from filtered media from subsequent passage from the experiment in (a). Error bars correspond to the 95% confidence interval from Poisson Distribution in the ddPCR analysis algorithm centered around the mean from each reading that contained >15 K droplets replicates. Additional virus infection kinetics measurements are shown in Supplementary Fig. S8. T-flask illustration from NIAID NIH BioArt Source (bioart.niaid.nih.gov/bioart/303).

We followed established serial propagation and blind passaging procedures using C6/36 Ae. albopictus cells and Aag2 Ae. aegypti cells to test for virus infection from the homogenates of the BZL lab strain (Fig. 5b), a colony strain from Poza Rica, Mexico (PZR, Supplementary Fig. S8a), and two African colony strains AWK and KIN (Supplementary Fig. S8b). We chose only these four mosquito strains because of available stocks of mosquitoes to generate ample lysates, and after 5-to-6 rounds of blind passaging, we were able to use RT-PCR and primers to consistently detect the stable infections of TMBTLV RNAs S1 and S2, the RdRP and Capsid genes of Verdadero virus, and a Guadeloupe Totivirus (GTTV) and FORMV amplicon. To demonstrate the broad capacity of mosquito cell infection by these viruses, we could also perform blind passaging infections in distinct U4.4 Ae. albopictus cells and CCL-125 Ae. aegypti cells (Supplementary Fig. S8d)

These results suggested we could establish stably infected mosquito cells as stocks of these mosquito viruses, albeit in mixtures of the viruses originally in the mosquito sample and at moderate virus copy levels (i.e., 20K–200K virus copies per mL of infectious media). Since there is no titer regime for these viruses and many other mosquito insect viruses^11,73–75, we followed the same procedures in those studies to use quantitative RT-qPCR and droplet digital PCR (ddPCR) as appropriate proxies to conduct infection kinetics in mosquito cells and virus tropism for TMBTLV, Totivirus, and FORMV in C6/36-NL cells than Aag2 cells. TMBTLV infection rates proceeded faster and to a greater extent in C6/36-NL cells than Aag2 cells (Fig. 5c and Supplementary Fig. 8d), likely because of the Dcr2 mutation in C6/36 cells that reduces antiviral RNAi and makes C6/36 cells the most common cell culture system to isolate viruses from insects^76,77. In contrast, FORMV exhibited efficient replication only in Aag2 cells, suggesting that this virus may have a preference for replicating in Ae. aegypti cells more than Ae. albopictus cells. TMBTLV and GTTV replicated in both cell types, although replication kinetics were a bit faster in C6/36 cells.

Next, we tested if these three viruses can also infect mammalian Vero E6 and Huh7.5 cells that are susceptible to mosquito-borne human viruses like ZIKV and DENV. Although the BZL mosquito homogenates inoculated into these mammalian cells only showed residual amounts of TMBTLV in the first blind passage, these mosquito viruses were unable to truly infect and replicate in the mammalian cells (Supplementary Fig. S9). However, the parallel virus infections in the C6/36 mosquito cells were robustly transmitted during each blind passage, and some CytoPathic Effects (CPE) were observed more frequently in the C6/36-NL line that has been known to already be persistently infected with other mosquito viruses³² (Supplementary Fig. S9).

Stably infected mosquito cells generate strong vsmRNA responses that exert gene silencing activity

We subjected our virus-infected C6/36 cells and Aag2 cells to small RNA sequencing to determine whether robust vsmRNA biogenesis would emerge in these cells. These C6/36 and Aag2 cell lines were obtained from other mosquito laboratories and already had known viruses persisting and generating small RNAs^32,78–80 (Fig. 6a, b). We were able to reconfirm the presence of Cell Fusing Agent Virus (CFAV) and PCLV in both of our ‘mock’ C6/36-NL and Aag2 cells³², as well as discover Sobemo-like viruses and densoviruses in our C6/36-NL and Aag2 cells, respectively. Furthermore, we compared the C6/36-ATCC line, which started out as a virus-free line because it was used for generating a tetravalent DENV vaccine⁸¹, and this line could also be readily infected like the C6/36-NL line. In fact, the mosquito viruses persisting in the C6/36-NL line were so infectious that they transferred to the C6/36-ATCC lines after prolonged culturing via routine biosafety cabinet work (Supplementary Fig. S10).

Fig. 6. Mosquito viruses infecting cell cultures can trigger an RNAi response by generating abundant vsmRNAs.

Bubble plots of vsmRNAs sequenced from (a) Ae. albopictus C6/36 cells and (b) Ae. aegypti Aag2 cells, either mock or stably infected after multiple blind passages of virus stocks from mosquito homogenates. Number of reads per million are reflected by bubble diameter, and color represents strand bias of reads, red is plus strand biased, blue is minus strand biased. Magenta boxes and dashed circles in (a) and (b) highlight specific samples inspected in coverage plots in (c) and (d). Coverage plots for four broadly-infecting viruses, TMBTLV, ANPHV, Guadeloupe Totivirus (GTTV) and FORMV in (c) Ae. albopictus C6/36 cells and (d) Ae. aegypti mosquitoes and Aag2 cells with the intact mosquito from females shown in the middle for comparison. Some vsRMNA patterns resemble and differ from the patterns in whole mosquitoes.

In addition to the preexisting viruses in the background-strain cell lines, these infected mosquito cells exhibited vsmRNAs from TMBTLV (from BZL mosquitoes), Anphevirus (from PZR mosquitoes), and FORMV (from AWK and KIN mosquitoes). The C6/36 cells generated a strong viral piRNA response to all three viruses with resemblance to the primarily plus-strand viral piRNA patterns observed in the cognate whole mosquito sample (Fig. 6c). The Dcr2 mutations in C6/36 cells prevent these cells from generating conventional siRNAs from a double-stranded RNA (dsRNA) intermediate made during virus replication⁷⁶, which could partially explain the lack of virus minus strand siRNAs.

The TMBTLV infection of Aag2 cells, which have an intact Dcr2 gene, showed a clear pattern of viral siRNAs processed from a dsRNA intermediate since minus and plus strand siRNAs evenly covered the entire virus genome (Fig. 6d). This intact antiviral RNAi response in Aag2 cells against TMBTLV explains the slower infection kinetics in Aag2 cells versus C6/36 cells (Fig. 5c). However, ANPHV and FORMV both did not generate antisense siRNAs in Aag2 cells but instead produced mainly plus strand piRNAs from all the viral genes except the RdRP gene, similar to the small RNA patterns in the whole mosquito (Fig. 6d). Lastly, vsmRNA patterns from TMBTLV and FORMV were observed in CCL-125 and U4.4 cells, affirming the reproducibility of these viruses in infecting multiple mosquito cells and triggering an RNAi response (Supplementary Fig. S11).

These mosquito cells stably infected by FORMV and TMBTLV also expressed massive amounts of viral long RNA transcripts despite also generating abundant vsmRNAs (Fig. 7a). This begs the question of whether these abundant vsmRNAs are capable of RNAi and gene silencing activity in mosquito cells? Although only two abundant satellite-sequence-derived mosquito piRNAs from Ae. aegypti were shown to directly silence luciferase reporters^82,83, no other study has yet reported mosquito vsmRNAs directly exhibiting gene silencing activity.

Fig. 7. Reporter assays demonstrating gene silencing by mosquito vsmRNAs.

a Design of Nano-luciferase (Nn-luc) reporter constructs with 3’ UTRs that contain a segment of FORMV or TMBTLV that is targeted by the vsmRNAs displayed in the coverage plots for Aag2 and C6/36-ATCC cells. The reporter constructs with virus segments inserted in the Sense (S) orientation have the same sequence as the Plus strand vsmRNAs (red bars in coverage plots), while the AntiSense (AS) oriented fragments can base-pair against these vsmRNAs. b One biological replicate (with technical triplicates) showing the first level normalization of expression of Nn-luc reporter constructs with virus segments transfected into Aag2 cells that either has been infected by AWK mosquito homogenate containing FORMV (Infct.) or has not been infected (Mock). c Relative reporter expression ratio between transfections into AWK-mosquito Infected cells divided by Mock cells for FORMV segments tested in Aag2 cells and C6/36 cells. The arrows from panel (b) direct the reader to the multiple normalizations of each reporter from biological quadruplicates shown in panel (c). For this FORMV reporter, 10 ng and 0.1 ng of Nn-luc reporter constructs transfected into Aag2 and C6/36 cells, respectively. d Similar multiple-normalized relative reporter expression ratios for biological triplicates of the TMBTLV segment reporters transfected into Mock Aag2 and C6/36 cells and cells stably infected from BZL-mosquito homogenates containing TMBTLV. For this TMBTLV reporter, 10 ng and 0.1 ng of Nn-luc reporter constructs transfected into Aag2 and C6/36 cells, respectively. Error bars represent standard deviations around the mean and P values are from two-tailed unpaired T-tests.

To answer this question, we constructed a series of Nano-luciferase (Nn-Luc) reporter plasmids that contained a segment of FORMV or TMBTLV in the reporter’s 3’UTR, in a similar format to reporter genes that we have used successfully to confirm transposon piRNA silencing activity in Drosophila cells^84,85. We picked single ~2.2kb-sized FORMV and TMBTLV segments that we could insert into Nn-Luc in either Sense orientation (S, same plus strand as the virus transcripts) or AntiSense orientation (AS, able to base-pair to the bulk of plus strand vsmRNAs displayed in Fig. 7a). We then transfected these Nn-Luc reporters and an internal control firefly luciferase plasmid into Aag2 and C636-ATCC cells that were either mock uninfected or stably-infected by FORMV and TMBTLV and assayed reporter levels 2 days later.

Luminescence readings from one biological replicate experiment in Fig. 7b show the FORMV reporter transfected in Aag2 cells. The Sense-oriented Nn-Luc reporters were expressed at similar levels between mock and infected cells, whereas the AntiSense (AS)-oriented Nn-Luc reporters were expressed at much lower levels in the virus-infected cells compared to the mock uninfected cells. After three biological replicates of these experiments were further normalized for each replicate to be compared together, we consistently observed strong repression of the AS-oriented reporter that can be targeted by the bulk of plus-strand vsmRNAs from FORMV stably infecting Aag2 cells (Fig. 7c). The reporter gene-silencing magnitude is lower in the infected C6/36-ATCC compared to infected Aag2 cells since much less Nn-Luc plasmid was titrated for the transfection so that the order of magnitude fewer vsmRNAs in C6/36-ATCC cells could accommodate the silencing of this reporter.

Cognate reporters based on the S1-RDRP fragment of TMBTLV transfected into mock or infected Aag2 and C6/36 cells similarly showed only the AS-oriented reporter was strongly silenced in infected cells (Fig. 7d). Although C6/36 cells mainly generating plus-strand vsmRNAs to TMBTLV would explain the AS-reporter silencing (Fig. 7a), the higher levels of minus-strand siRNAs generated in Aag2 cells did not exert visible reporter gene silencing of the Sense-oriented reporter (Fig. 7d). Perhaps the massive expression of the plus-strand viral S1-RDRP transcript (Fig. 7a) could be interfering with the AS-oriented reporters’ expression.

Discussion

We conclude that mosquito cells stably infected with FORMV and TMBTLV generate abundant vsmRNAs and exert a reporter gene silencing response that is indicative of viral RNA sequence base-pairing to the AS-oriented virus segment in the reporter. Now that we have established this viral segment gene silencing reporter in mosquito cells, we open the door to future functional genomics experiments to quantify this viral silencing capacity by vsmRNAs versus mosquito-encoded miRNAs and endogenous siRNAs. Additional virology experiments beyond the scope of study will explore whether the vsmRNAs and viral mRNA transcripts persist at high levels in the stably infected mosquito cells, which might be due to the possibility of virus-encoded suppressors of RNAi seen in Drosophila viruses and arboviruses^67,86–91 that may also be encoded by FORMV and TMBTLV.

We acknowledge that our transcriptomics study may still carry the inherent challenge of resolving bona fide viruses from novel EVEs that would require additional whole genome sequencing of the surveyed mosquitoes to detect the unknown EVEs^12–19. Therefore, we discuss extensively in the Supplementary Text Discussion document with diagrams SDF-4 and SDF-5 to explain how we mitigate this concern by conducting the virus isolation procedures for several of the viruses uncovered by our small RNA genomics survey, as shown in the results of Fig. 5 and Supplementary Fig. S8. Only a true virus would be transferred from the filtered mosquito lysate to mosquito cell cultures, while EVEs that are just genome-integrate fragments of a virus would not be infectious.

We present a comprehensive global mosquito virus small RNA survey that demonstrates frequent and robust small RNA responses to persisting viruses within wild Ae. aegypti mosquitoes and in recently established colonies. We analyzed new small and long RNA sequencing datasets alongside previously published small RNA datasets¹¹ to yield a snapshot in time of the geographic distribution of mosquito viruses. Our analysis shows an instance of the global reach of PCLV (and to a lesser extent HTV) infecting Ae. aegypti, but geographical boundaries of mosquito viruses based on small RNA profiles did exist recently, such as ANPHV/ PCLV-HTV delineations in the Americas and African-specific viruses like FORMV. By comparing viral long RNAs to small RNAs, we uncovered new insights into the dynamics of virus replication intermediates that can evade RNAi interactions, mechanisms which we can explore in the future by infecting these viruses into mosquito cell cultures while knocking down mosquito host genes.

Climate change is expanding the ecological niches for mosquitoes like Ae. aegypti, which has now invaded further into North America and Europe⁵⁶. Broadening our small RNA genomics surveys to include animal captures in new locations would illuminate the potential disease vectoring threat that these invading mosquitoes would pose to human populations⁹². Thus, it is surprising to observe the first case of DENV small RNAs in a wild mosquito isolate (Fig. 2d). It remains an open question whether wild mosquitoes displaying arboviral small RNAs are still actively infectious or if the RNAi response is now suppressing virus transmission. Only sequencing approaches, not RT-PCR, can discover new viruses and RNAi responses in mosquitoes, and our study demonstrates the utility of this approach to add to existing arbovirus surveillance programs.

Previous studies on genetic variations causing lower ZIKV vector competence in Ae. aegypti formosus compared to Ae. aegypti aegypti were done prior to our new results on FORMV persisting and being transmitted vertically in these African colonies⁴⁰. It is possible that the formosus-aegypti crosses conducted in the previous study may have also transmitted FORMV, and insect virus interactions with arboviruses are documented. For example, co-infection with CFAV reduces DENV and ZIKV replication in cell culture and transmission in mosquitoes^93–95. Our next priority will be to test how these mosquito viruses interact with DENV and ZIKV in co-infection experiments.

An open question is whether persistent mosquito viruses can be transmitted from the mosquito to the bitten animal during blood feeding. For instance, it is unclear if massive vertical virus transmission of TMBTLV via maternal contribution to the eggs (Fig. 4b) would reflect a meaningful reservoir of viruses in the mosquito salivary glands and proboscis. On the other hand, LTLV had very high levels of small RNAs in the midgut (Fig. 2b, Supplementary Fig. S5a), suggesting this virus could likely mix with an initial bloodmeal and be transmitted during subsequent blood feedings. Notably, mosquito viruses like TMBTLV are related to plant viruses, suggesting that mosquito feeding on plant sap could be a common modality of virus transmission^63–65, and perhaps these viruses are suppressing RNAi with factors related to P19 found in plant tombusviruses⁶¹.

In addition, these mosquito viruses may also be useful agents for mosquito transduction and control in a similar vein as insect densoviruses^96,97. Densoviruses have been engineered to carry reporter genes and toxic gene knockdown cassettes as prototype genetic tools to manipulate mosquitoes, but technical challenges must be overcome to realize the potential of densoviruses as mosquito genetic tools. We observed that, like the insect viruses investigated in this study, densoviruses also engage with the mosquito RNAi pathway to generate vsmRNAs (Fig. 6b and Ma, et al.³²). Rivaling densovirus in genome compactness is TMBTLV at just ~4.5 kb across two RNA segments. We are still characterizing the viral genomic RNAs’ 5’ and 3’ ends to build and test infectious clones, but there is potential to develop new mosquito genetic tools to be garnered from such a survey.

Previous studies have tried to harness RNAi to control mosquito populations and curtail arbovirus transmission^98–103, but dsRNA delivery by injection has a low throughput, and RNAi by dsRNA ingestion is also limited in mosquitoes¹⁰⁴. Mosquito viruses like FORMV and TMBTLV that can generate dsRNA in mosquitoes and cells may become a potent future vector control tool, but our future studies will explore how long viral transcripts can either escape the RNAi pathways or be shunted in viral piRNAs versus viral siRNAs. This study now demonstrates that a reporter-gene platform can interrogate virus-directed gene silencing that will enable future functional genomics screens of mosquito RNAi factors influencing the vsmRNA response (Fig. 7).

It is possible that some of the viruses detected in colonized mosquitoes could result from a lab-acquired infection when multiple colonies share the same insectary space. Colonies that have been perpetuated for many generations may begin to reflect the environment in which they are raised instead of the locale from which they originate and are meant to represent. For instance, we detected the American-prevalent ANPHV and CFAV in the Lambrechts lab’s Lope strain and the McBride lab’s ZIK strain, respectively. These African-originating strains had the most generations of lab propagation (>15 generations) compared to the other African colonies (average ~8 generations, Fig. 3A).

It is practical for academic lab mosquito strains to be clean of persistent viruses that would complicate genetic analyses. However, to better model arbovirus vector competence in the wild where human populations are affected, we need to study further the impact of persistent viruses that generate abundant vsmRNAs in mosquitoes. The infectivity of the mosquito viruses into cell cultures was surprising, considering the apparent vsmRNA responses exhibited by the mosquito animals and the cell cultures. We did have occasional issues with mosquito viruses transferring between culture flasks by possible aerosolization within the biosafety cabinet during the handling of multiple mosquito homogenates (Fig. 6). Simple routine culturing of C6/36-ATCC cells that started out virus-free became virus infected by potential transfer via the pipettor aerosols (Supplementary Fig. S10). This infectivity could explain the persistence and widespread detection of these viruses in so many of the mosquito samples profiled in this study.

Methods

All unique/stable reagents generated in this study are available from the lead contact. Material transfer agreements with Boston University may apply. Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Nelson Lau (nclau@bu.edu).

Construction of small and total RNA libraries from whole mosquitoes and cells

Total RNA was extracted from whole frozen mosquitoes or mosquitoes preserved in RNA protection solutions. In most cases, 5-10 mosquitoes were used. Numbers of mosquitoes in the pool are indicated in Supplementary Data 1. Zirconium beads (3.0 mm) in a bead beater were used to homogenize mosquitoes before proceeding with NEB Monarch Total RNA Miniprep Kit (NEB #T2010). The on-column DNase I treatment was performed during RNA extraction. All mosquito total RNA extractions were checked on a NanoDrop beforehand to ensure RNA quality A260/280 ratios no lower than 1.5 and generally above 1.8.

Small RNA libraries were made using NEBNext Small RNA Library Prep Set (NEB #E7330) with up to 5 µg of RNA input. During library amplification, samples were put through up to 25 total PCR cycles. Total RNA libraries were made using Zymo-Seq RiboFree Total RNA Library Kit using up to 250 ng RNA input, and following the manufacturer’s protocol. All libraries were checked on an Agilent Bioanalyzer 2100 using either DNA 1000 or High Sensitivity DNA kit and sequenced at the BUMC Microarray and Sequencing Resource on an Illumina Next Seq 2000 using P3 flow cells for 50SE and 50PE reads for small RNA and total RNA libraries, respectively.

Running VirusDetect with TPA and MAG database modifications

The source code for VirusDetect Version 1.7⁵² was downloaded from the GitHub repository (https://github.com/kentnf/VirusDetect). Additionally, the preprocessed VRL virus database Version 248 was obtained from the VirusDetect webpage (http://bioinfo.bti.cornell.edu/ftp/program/VirusDetect/virus_database/v248/). The source code and the corresponding database were deployed on the Shared Computing Cluster of Boston University. To meet VirusDetect requirements, the following packages were installed on the cluster: perl version 5.28.1, bioperl version 1.7.2, and python3 version 3.8.10.

The Third Party Annotation (TPA) GenBank files were sourced from the NCBI database (https://ftp.ncbi.nlm.nih.gov/tpa/tsa/). An in-house R pipeline was utilized to extract GenBank records, preserving 45,601 entries with the taxonomy ‘VIRUS’ for downstream processing. For each entry, both the genome sequence and corresponding protein sequences were retrieved from the NCBI database using the ACCESSION ID and saved in database files formatted for VirusDetect. Two id-mapping files were also generated following VirusDetect’s instructions. In addition to the GBVRL and TPA datasets, partial virus sequences in GenBank were annotated as Metagenome Assembled Genome (MAG). All MAG records downloaded from GenBank underwent the same processing steps as the TPA database. The TPA version was downloaded 2022-09-06, and the MAG version was downloaded on 2023-03-11. A total of 65,167 entries were retained to construct the modified VirusDetect database. Default parameters were used when running the VirusDetect pipeline on samples.

Sequencing and bioinformatics analysis of small and long RNAs from mosquitoes

RNA Libraries were selected to run through the Mosquito Small RNA Genomics (MSRG) pipeline^32,51 if they had above 10 million reads before adaptor removal. After adaptor removal in MSRG, reads greater than 45nt were filtered out. Outputs from MSRG include the alignment to a curated virus database. This includes breakdowns (in reads per million) of reads coming from siRNAs (18–23 nt) and piRNAs (24–35 nt), which strand reads map to, and a ratio of peaks to the average distance between peaks (used as a metric of coverage across viral sequence).

Some of the libraries in the Olmo et al. study¹¹ were made using a modified version of the NEBNext Multiplex Small RNA Library Prep Set where a random 6-mer was attached to the 5’ adapter. A Cutadapt trimming step in MSRG was adapted to remove the first six bases of each read. This modified pipeline was then used to run MSRG on these samples.

Total RNA libraries were sequenced as 50PE reads, but were processed to look like small RNA reads so that they could also be run using the MSRG pipeline. All reads were trimmed to 35nt, R1 was reverse complemented to be in the same orientation as R2, and then the reverse complemented R1 was merged with R2. The resulting FASTQ file was then run through the normal MSRG pipeline. Only the last 15nt at the 3’ ends of the 50nt reads from longer RNA sequencing were trimmed to yield 35nt reads so that the MSRG code can use these reads to generate RNA coverage plots that properly depict the virus RNA genome coverage based on the 5’ ends of the reads. The minimal sequence trimming and the stringent mapping of the trimmed long RNA reads in the MSRG pipeline provide the expression coverage of the virus RNA genome before it is processed into vsmRNAs.

Viruses were only considered to be present in a sample if it had at least 10 reads per million (RPM) and a coverage ratio greater than 0.75. Additionally, a strand bias score was calculated for each virus/sample pair by taking the ratio of reads mapping to the top versus bottom strand. For many of the field collected samples, the small pools of 5–10 mosquitoes are the standard operating procedure of the Vector Control districts in California and Florida, and these pools all come from a single site representing the local population of Ae. aegypti, which is a cosmopolitan mosquito with a short dispersion range^105,106. Thus, the sampling of viruses with these read counts cutoffs provide a good threshold for reflecting reasonably high viral RNA levels that can be constrained to the location of the field collection even when multiple viruses are detected.

For plots made using R/RStudio, packages used include: ggplot2, tidyverse, and ggpubr. In the bubble plot, log₁₀ of the strand bias score was mapped to the color, and log₁₀ of the RPM was mapped to the size. For the strand bias, any value greater than 1 (a greater than 10-fold strand bias) was squished to be the “max” color on the scale. For the scatterplots comparing the small and total RNAs, the 10 RPM ratio filter was ignored in order to completely compare the two libraries. Sample/virus pairs were kept if at least one of the two libraries was at least 10 RPM.

The phasing and ping-pong piRNA biogenesis signatures were calculated from algorithms within the (MSRG) pipeline^32,51 and were applied to the specific BED files of vsmRNAs. The dataframe and R scripts and virus genome lists for generating the bubble plots are enclosed in the Supplementary Data Zip File in the manuscript supplement.

We used BLAST and BLAT to define EVEs against the mosquito viruses and determined which putative virus candidates to be removed because of mis-annotations from an existing transposon, an approach similarly applied by other studies. Virus sequences were subjected to default parameters of BLAT or BLASTN (using somewhat similar sequences parameters) against the Ae. aegypti and Ae. albopictus genomes requiring a match of at least a contiguous >100 bp of sequence to accept as an EVE. When the query results in multiple genomic loci matches like for the putative Guato and TO virus entries, we interpreted these as mis-annotations of an existing transposon (See Supplementary Text Discussion).

Cell culture conditions and infections with mosquito viruses

C6/36 and Aag2-TC mosquito cell cultures were propagated in DMEM (Gibco) supplemented with 10% FBS and 1% Tryptose Phosphate Broth (TPB) and maintained at 28° C and 5% CO₂. HUH7.5 and Vero E6 TC cell cultures were propagated in DMEM (Gibco) supplemented with 10% FBS and 1% P/S (Penicillin/Streptomycin) and maintained at 37° C and 5% CO₂. The mammalian cell cultures were split by washing with 1X PBS and 0.25% Trypsin EDTA.

The mosquitoes received from Awka (Nigeria), Kintampo (Ghana), Pozo Rica (Mexico), and Banzon Research (USA) were used to infect the cell lines with viruses. Mosquitoes were homogenized in 1.5 mL DMEM with zirconium beads (3.0 mm) in a bead beater. The homogenate was passed through a 0.45 µm pore size filter and added into a T-25 flask containing ~90% cell confluency in 5 ml of media in T-25 flasks. On the first day of infection only, a cocktail of primary cell cultures antibiotics from InvivoGen was only added to the infection media to mitigate concerns of microbes in the mosquito homogenate.

For mosquito cell blind passage, the medium was collected from the infected cells, passed through the 0.45 µm pore size filter, and added into a new T-25 flask containing ~90% cell confluency. For serial passage, the cells infected by homogenizing the mosquitoes were split into a new flask with a fresh medium. This process was repeated every 7th day for both blind and serial passage. The cells collected on the 7th day were used for total RNA extraction to confirm the virus infection.

Mosquitos infected with viruses (Formosus, Totivirus, and Tombus-like Virus) were homogenized in Gibco DMEM (1X) and were added to both C6/36-ATCC and C6/36-NL cells to propagate viral particles into the media. The infected media was filtered through 0.45 µm pore size filter and was added to T25 flasks containing Vero E6 and Huh7.5 cells. Once media was added, cells were imaged at 10X magnification on an EVOS M5000 microscope. Cells were incubated for 3 days before being imaged again for any potential cytopathic effects incurred by viral propagation. The current infected media was aspirated, and cells were washed twice with 1X PBS. Fresh, non-viral media was added onto the cells. The cells were incubated for an additional 3 days before being imaged once more. Media was collected by filtering through a 0.45 μm pore size filter. Cells were washed twice more with 1X PBS before being extracted using the NEB Total RNA Extraction kit by adding 1X DNA/RNA Protection Reagent and RNA Lysis buffer to remove the cells from the flasks. Total RNA was extracted using the same kit, and RT-PCR assays were executed to detect any presence of viruses in cells incubated with viral media. For each following week, media collected on the 6^th day of each passage were added to a new passage of Vero and Huh7.5 cells instead of using viral media from C6/36 cells.

RNA extraction and RT-qPCR analysis of virus amplicons

Total RNA was extracted with Monarch Total RNA Miniprep Kit (NEB) following the manufacturer’s instructions, including the DNase I treatment step. For RT-PCR, 1 µg RNA was used for first-strand cDNA preparation by using ProtoScript II Reverse Transcriptase kit followed by Phusion High-Fidelity DNA Polymerase (NEB). RT-qPCR was performed with LunaScript RT Supermix (NEB) in a Bio-Rad CFX Opus 96 Real-Time PCR System. The housekeeping genes, Ae. aegypti actin (GenBank accession XM_001659913) and Ae. albopictus actin (GenBank accession XM_019702203), were used as reference genes to normalize the target gene expression by 2^(-∆∆Ct) methods. All oligonucleotide primer sequences against virus elements are listed in Supplementary Data 5.

Virus infection kinetics

The virus infection kinetics of TMBTLV, FORMV, and Totivirus were measured in C6/36 and Aag2-TC cell lines. The cells were infected with 5 mL virus stock medium in the T-25 flask. The cells were collected from the flask on the 1st, 3rd, 5th, 7th, and 10th day post-infection. At each time point, the cells were resuspended in the medium using a pipette, and one mL of the medium containing the cells was replaced with a fresh medium. The cells were collected and processed from one mL for the total RNA extraction. Subsequently, the virus quantification was done by RT-qPCR, as described above. This experiment was repeated in three independent biological replicates for each time point. Virus particles were also quantified in the virus stock medium by digital-droplet PCR on the Bio-Rad QX-200 instrument.

Luciferase reporter assays

Segment 1 of FORMV (nt 22–3897) and TMTLV (nt 33–2119) was amplified using Phusion High-Fidelity DNA Polymerase. Segments were cloned into the PCR-4 Blunt TOPO vector. The corresponding insert was then re-amplified from the TOPO construct using primers (Supplementary Data 4) flanking the cloning site, with KpnI adapter sequences added at the 5′ end of each primer. The resulting amplicons were cloned into the 3′ untranslated region (UTR) of a luciferase reporter plasmid under the control of the IE1 promoter (derived from baculovirus) at the KpnI restriction site in both sense and antisense orientations. Similarly, segment S1 of TMBTLV was amplified from their respective TOPO constructs using the same strategy.

C6/36 and Aag-2 TC mosquito cell lines were stably infected with AWK and BZL viruses. Mock (uninfected) cells were included as controls. Cells were seeded into 24-well plates, and various concentrations of recombinant plasmids were transfected. In C6/36 cells and Aag2 cells, the optimal amount of plasmid DNA for transfection was titrated to 0.1 ng and 10 ng, respectively. Alongside each treatment, 400 ng of NN-FF plasmid was co-transfected for normalization. Fugene HD (Promega) was used as the transfecting reagent.

Plasmids NN-Luc-S1-FORMV and NN-Luc-S1-TMTLV, containing viral inserts in both orientations, were transfected into both virus-infected and mock-infected cells. After 48 h, luminescence was measured using ONE-Glo reagent and S&G substrate. Luminescence was read on a BioTek luminometer and raw luminescence values of the recombinant constructs were first normalized to the corresponding NN-FF control. A second normalization was performed using luminescence from mock-transfected cells. Wilcoxen tests were used to assess statistical significance between sense and antisense constructs.

Reporting summary

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

Author contributions

Conceptualization, N.C.L., S.G., R.S., A.E.W., and Z.Z.; Methodology/Investigation, N.C.L., S.G., R.S., A.C.-V., M.K-K., A.E.W., Z.Z., and G.D.; Formal Analysis, S.G., N.C.L., and R.S.; Data Curation/Software, C.Z., S.G., and Z.Z.; Writing—Original Draft, N.C.L., S.G., and R.S.; Writing – Review & Editing, N.C.L., S.G., R.S., A.E.W., L.L., N.H.R., C.S.M., and Z.Z.; Visualization, N.C.L., S.G., R.S., and Z.Z.; Funding Acquisition, N.C.L., L.L., K.E.O., E.C., C.S.M., M.A.Y., D.E.B., and M.D.; Sample Contribution & Resources, A.E.W., I.S.-V., N.H.R., A.G.-S., D.E.B., J.M., S.S.W., A.C., T.R., M.S., A.B., J.A., O.B.A., D.A., W.-L.L., C-H.C., C.V., C.G.A., A.P., T.M., B.H.C., D.M.W., D.S., M.A.Y., A.L.C.-S., M.D., A.B., L.L., C.S.M, K.E.O., and E.C.

Peer review

Peer review information

Nature Communications thanks the anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

All sequencing data produced and generated by this study is available on Sequencing Read Archive (SRA) under BioProject PRJNA1104658 [https://www.ncbi.nlm.nih.gov/sra?linkname=bioproject_sra_all&from_uid=1104658] and under GEO Series GSE309873 and GSE309876. See Supplementary Data 1 and Supplementary Data 2 for specific BioSample and SRA accessions. SRA accessions for publicly available datasets used in this study can be found in Supplementary Data 1 tab c. Source data are provided with this paper.

Code availability

The MSRG pipeline code can be found on the Github repository: https://github.com/laulabbumc/MosquitoSmallRNA. Additional R code and dataframes are packaged in the Source Data Zip File in the manuscript supplement. The VirusDetect program can be found at the Github repository: https://github.com/kentnf/VirusDetect.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-71964-1.