Ethical approval was obtained from the Kenya Medical Research Institute (KEMRI) Scientific and Ethics Review Unit (SERU) under protocol number KEMRI/SERU/CCR/4702 and WRAIR# 3101. Permission to conduct the study was granted by the National Council for Science, Technology, and Innovation (NACOSTI).

The study was carried out in four study areas, Kilifi County (Malindi) (3.2192°S, 40.1169°E), Kwale (4.1730°S, 39.4520°E), Mombasa (4.0435°S, 39.6682°E), and Isiolo county (0.3546°N, 37.5822°E) ( Figure 1 ).

Sampling of adult mosquitoes was conducted between 18^th June and 10^th December 2023. Mosquitoes were collected using CDC miniature light traps (Model 512, John Hock Co., Gainesville, Florida, USA). Twelve (12) traps baited with carbonated dry ice (CO2) were deployed overnight (6pm-6am) in favorable habitat, including animal sheds and termite mounds. The samples were linked to the sites by geo-coding using a GPS. Captured mosquitoes were temporarily immobilized using triethylamine, preserved in liquid nitrogen, and transported to the laboratory at the Kenya Medical Research Institute in Kisumu, where they were stored at-80˚C until further processing. The mosquitoes were identified morphologically to species under dissecting microscope using taxonomy keys, including Edwards (1941) (15), Harbach (1988) (16) and Jupp (1986) (17). The identified mosquitoes were pooled in groups of 1 to 25 samples based on species, sex and collection site and stored at -80 degrees Celsius. In this study, mosquitoes were not pooled by abdominal status (i.e., unfed, blood-fed, gravid), this decision was based on the primary objective of virome detection and genomic characterization.

One thousand five hundred ninety-six (1,596) mosquito pools were homogenized using a Mini-Beadruptor-16 (Biospec, Bartlesville, OK, USA) in 1000 µL of homogenization media (minimum essential media supplemented with 15% fetal Bovine Serum (FBS) (Gibco by Life Technologies, Grand Island, NY, USA), 2% L-glutamine (Sigma, Aldrich)and 2% antibiotic/antimycotic (Gibco by Life Technologies, Grand Island, NY, USA.) with zirconium beads (2.0 mm diameter) for 40 seconds. Subsequently, the homogenate was centrifuged at 14,000 rpm for 15 minutes at 4°C using a benchtop centrifuge (Eppendorf, USA). The supernatant was collected and inoculated into Vero cells while maintaining cold chain throughout the process.

Virus isolation was performed in Vero (African green monkey kidney) cells (E6). Briefly, Vero cells were grown overnight at 37°C and 5% CO₂ in minimum essential medium supplemented with 2% glutamine, 2% penicillin/streptomycin/amphotericin, 10% fetal bovine serum, and 7.5% NaHCO₃ in 24-well plates (Corning, Incorporated). At 80% confluence, a 50 µL aliquot of the clarified supernatant from individual pools was inoculated into the wells. The plates were incubated for 1 hour in a humidified incubator at 37°C and 5% CO₂ with gentle rocking of the plates every 15 minutes for virus adsorption. Following incubation, 1 mL of maintenance medium, comprising minimum essential medium supplemented with 2% glutamine, 2% penicillin/streptomycin/amphotericin, 2% fetal bovine serum, and 7.5% NaHCO₃, was added. The plates were then cultured at 37°C and 5% CO₂ and monitored daily for cytopathic effects (CPE) for 14 days. Cultures exhibiting CPE were harvested and further passaged by inoculating onto fresh monolayers of Vero cells (CCL 81™) in 25-cm³ cell culture flasks. After three successive passages, the supernatants of virus-infected Vero cell cultures exhibiting cytopathic effect of approximately 70% were harvested from the flasks for virus identification through next-generation sequencing.

Viral particles from CPE-positive cultures were recovered through 0.22µm filters (Millipore, Merck). From an aliquot of 140 μl of the supernatant, viral RNA was extracted using the QIAamp Viral RNA Mini Kit (Qiagen, Hilden, Germany) and eluted in one step with 60 μl of elution buffer. Paired-end libraries for high-throughput sequencing were prepared using Illumina Ribo-Zero Plus rRNA Depletion (Illumina, USA). The libraries were prepared following the manufacturer’s recommended protocol. The final library was denatured with NaOH and diluted further to 10pM before loading onto the Illumina MiSeq sequencing machine. Sequencing was performed using MiSeq Reagent V3 (Illumina, USA), in a 300-paired end cycle sequencing format.

Initial analysis was performed using the CZ-ID platform, an integrated pipeline that offers quality control, de-hosting, duplicate removal, assembly, and viral identification capabilities. After this initial processing, the de-hosted sequence reads were retrieved for further analysis. To validate CZ-ID pipeline results, PrinseqLite v0.20.4 tool was used to filter low-quality reads and remove adapters on command line. De novo sequence assembly was conducted using MEGAHIT v1.2.9 (18) where Ndumu virus contigs were recovered and compared to CZ-ID pipeline results. This was followed by mapping the reads back to the generated contigs using BWA to produce SAM files. The subsequent generation of BAM files, along with sorting, indexing, and filtering of contigs that did not meet the specified read mapping thresholds, was handled using Samtools v1.20. Only contigs with an average depth of coverage of ≥10 and a length of ≥500 bp were retained for further analysis. These contigs were first compared against a local version of NCBI viral database using Diamond v2.0.4. To ensure specificity, putative viral contigs were further compared to the entire non-redundant protein database (nr), to exclude any non-viral contigs. A stringent e-value threshold of 1e-5 was employed throughout the homology searches to minimize false-positive hits.

To describe the identified viruses in an evolutionary context, publicly available viruses belonging to these different groups, and more specifically those closely related to the viral strains obtained in the current study were downloaded and used as reference sequences in the reconstruction of phylogenetic trees. Closely related gene sequences were retrieved from NCBI viral database and used as reference sequences in reconstructing the phylogenetic relationship of the viral sequences. The combined set of sequences were aligned using MUSCLE software with default parameters (max iterations = 16) embedded in Molecular Evolutionary Genetics Analysis v.7.0 (MEGA7) (19) platform. The aligned sequences were edited using the Bioedit tool and maximum likelihood phylogenetic analysis carried out using IQ-TREE v1.6.12. The best model (GTR+G4 (General Time Reversible + Gamma)) and tree search was performed simultaneously based on 1000 bootstrap estimates and approximate likelihood ratio test (aLRT).

COI gene amplification and sequencing were conducted to confirm the morphological identification of mosquito species in each pool. DNA extraction was performed on the bulk pools of the crude homogenates using a QIAamp DNA extraction kit (Qiagen, Germany), according to the manufacturer’s instructions. Two extraction blanks were included during the extraction process and subsequently used during PCR and sequencing. The extracted DNA was quantified using a Qubit fluorometer 2.0. Amplification of the COI gene was then carried out on <1µg of extracted DNA, using the universal pair of primers for metazoan invertebrates LCO1490/HCO2198 (20). These primers amplify an approximately 710-bp region of the COI gene of arthropod vectors. COI amplicons were generated from a 25µL PCR containing 12.5 µl AmpliTaq Gold 360 master mix (Applied Biosystems, USA), 9.5µl DNase/RNase-free water, and 0.5µl each of the forward and reverse primers at 25µM. The PCR cycling conditions were set as follows; initial denaturation at 95°C for 10 min, 35 cycles of 95°C for 30 s, 49°C for 30 s, 72°C for 30 s, and a final extension of 72°C for 10 min. The COI amplicons were first purified using AMPure XP beads (Beckman Coulter, USA). The purified products were quantified using a Qubit dsDNA HS assay kit (Invitrogen, USA) with a Qubit fluorometer 2.0. Based on the concentration of the quantified products, the volume of PCR products that yielded 200 fmol was determined and used as starting material for MinION library preparation. Library preparation was carried out using a ligation sequencing kit (SQK-LSK114), following the manufacturers protocol with the exclusion of the DNA fragmentation step. Briefly, 200 fmol of the purified products were end repaired using a NEBNext Ultra II end repair and dA-tailing module (New England Biolabs [NEB], UK). The end-repaired DNA for each sample was individually barcoded using Native Barcoding Expansion 1-12 (EXP-NBD114), which was achieved with the use of NEB Blunt/TA ligase master mix (NEB, UK). An equal amount from each of the 200-fmol barcoded libraries was combined into a single pool, which was then purified with AMPure XP beads (Beckman Coulter, USA). Adapter ligation of the purified library was done with NEBNext quick ligation module (NEB, UK) and the libraries were further purified using AMPure XP beads, with a final wash of the beads being carried out using short fragment buffer (SFB) provided with the SQK-LSK114 kit. The final library was loaded onto the flow cell (FLO-MIN106D) and sequenced using the workflows provided in the MinKNOW software. Raw reads were processed in real-time using the EPI2ME platform (Run on cloud), which performed base-calling with Guppy (v4.2.2), quality filtering (minimum Q-score 7), and taxonomic classification using the ‘What’s in my Pot’ (WIMP) workflow. Taxonomic assignments were validated by BLAST searches against the NCBI nucleotide database and cross-checked with the Barcode of Life Data System (BOLD) for species confirmation. Only sequences with ≥98% identity and ≥90% query coverage were considered reliable identifications.

To assess the genetic distance between different strains of Ndumu Virus and alphaviruses, we used the BioPython software v1.85 to calculate pairwise genetic distances based on the nucleotide sequences. The distances were computed using the Kimura 2-parameter model, p-distance. A pairwise distance matrix was constructed from the calculated genetic distances. The matrix was visualized as a heatmap using Seaborn (v0.11.0) in Python.

A total of one thousand five hundred ninety-six (1,596) mosquito pools were processed for cell culture, three pools originating from Malindi and Kwale exhibited cytopathic effects (CPE) on day 6 post inoculation ( Table 1 ). Ndumu virus was confirmed in Culex annulioris, Mn. africana and Culex univittatus mosquito species. The remaining 1,593 pools did not show any CPE across four blind passages and were considered negative using this approach ( Supplementary Material S1 ).

Phylogenetic analysis was conducted to characterize the Ndumu virus (NDUV) strains isolated from mosquito pools in this study and to determine their evolutionary relationships with other alphaviruses. The three Kenyan NDUV isolates formed a well-supported monophyletic clade closely related to a strain previously isolated from domestic pigs in Uganda, suggesting regional circulation and evolutionary divergence. The broader analysis confirmed that the NDUV strains clustered distinctly from Middelburg virus and other members of the Semliki Forest antigenic complex, while remaining within the Old World alphavirus clade ( Figure 2 ).

The COI sequence analysis provided a molecular identity for the mosquito specimens. These results were then compared to the species identity provided using morphologically based taxonomic keys ( Table 2 ). The results obtained showed that all the species that were collected in this study had concurring morphological and molecular identities.

To determine the degree of similarity and divergence, a genetic heat map was generated to visually represent pairwise genetic distances between the isolates and reference sequences ( Figure 3 ).

Heatmap analysis based on pairwise p-distances of full genome sequences revealed a clear clustering pattern among Ndumu virus (NDUV) isolates and distinct separation from other alphaviruses. The intra-group distances among the NDUV isolates ranged from 0.01 to 0.15, indicating high sequence similarity, with most isolates showing minimal divergence (0.01–0.06), except for one JX644171.1 notably divergent isolate with a distance of 0.15. In contrast, distances between the NDUV isolates and other alphaviruses such as Middelburg virus, Getah virus, and Chikungunya virus (CHIKV) were substantially higher, ranging from 0.400 to 0.430. Specifically, Middelburg virus showed distances of 0.400–0.410, while Getah virus and CHIKV ranged between 0.410–0.430. Consistent with other alphaviruses, the genome organization of NDUV retained the canonical arrangement of non-structural and structural proteins, with conserved functional domains, further supporting the biological stability of NDUV despite its genetic distinctiveness ( Figure 4 ).

This study provides strong genomic, phylogenetic, and functional evidence that Ndumu virus is an emerging alphavirus with the ability to infect both mosquitoes and mammalian cells. While no human cases have been reported, its phylogenetic proximity to other pathogenic alphaviruses and its ability to cause CPE suggest a potential public health concern. Further investigations, including serological studies, in vivo infection models, and vector competence assays, are needed to assess its zoonotic risk and transmission dynamics.