Genomes from Ethiopia’s first Marburg outbreak reveal post-spillover human transmission

In November 2025, Ethiopia reported its first outbreak of Marburg virus disease, with a case fatality rate of 64%. A new genomic analysis, published in Nature Microbiology, reports seven near-complete Marburg virus genomes recovered from seven early cases in that outbreak. The sequences show that the outbreak lineage belonged to the MARV.A.1 clade, a group closely related to genomes from historical human cases and from bats. High nucleotide identity across the seven genomes points to human-to-human transmission following a zoonotic spillover event, in which the virus jumped from an animal reservoir into people and then spread between them. The authors argue that the findings underline the value of genomic surveillance for outbreak detection, response and preparedness.

The work was carried out by researchers from the Ethiopian Public Health Institute (EPHI) and the Federal Ministry of Health, together with long-standing collaborators at the Amsterdam Institute for Global Health and Development, Amsterdam University and McMaster University, through the EpiGen-Ethiopia project, which is building scalable capacity for pathogen genomic surveillance in Ethiopia for emerging and re-emerging infectious diseases. The study was approved by the EPHI Institutional Review Board (EPHI-IRB-633-2025). Given the emergency situation, verbal consent was received from all patients included in the report, and all collected samples were anonymized to prevent the identification of any individual. The researchers were involved in all stages of the work, including study design, epidemiological investigation, community engagement, data collection, analysis and manuscript preparation.

The samples at the heart of the study were blood specimens collected from patients suspected of having viral haemorrhagic fever, presenting with symptoms such as high fever, severe headache, muscle pain, fatigue, nausea, vomiting, diarrhoea and bleeding. The samples were transported under standard operating procedures for infectious material to the EPHI in Addis Ababa. There, at the Integrated Pathogen Genomics and Bioinformatics Facility for Surveillance and Research, they were inactivated inside a fixed-film isolator using 100% ethanol at five times the sample volume. RNA was then extracted with a Qiagen kit inside a class III biosafety cabinet, following the manufacturer’s instructions, with lysing and purification buffers and a final elution step. The extracted RNA was screened by multiplex RT–PCR for dengue, Zika, Chikungunya, Yellow Fever, Ebola and Marburg viruses.

Of the specimens screened, 14 were RT–PCR-positive for Marburg virus. Not all of these yielded usable sequence data: two genomes had less than 52% breadth of coverage and five samples were unavailable for processing, leaving seven samples that underwent whole-genome sequencing. Library preparation used the Illumina RNA Prep with Enrichment workflow with the Viral Surveillance Panel 2. In this workflow, RNA was reverse transcribed to cDNA, enzymatically fragmented and amplified to add indexes and adaptor sequences. Library concentrations were measured with a Qubit 1× dsDNA High Sensitivity assay before enrichment, and target enrichment was performed by hybridization with capture probes from the Viral Surveillance Panel 2, followed by capture with Streptavidin Magnetic Beads 3, elution, amplification and purification with Illumina purification beads.

The enriched libraries were normalized, pooled in equimolar concentrations and diluted to the appropriate loading concentration for sequencing. Two libraries were diluted with NovaSeq X series Pre-Load Buffer to a final concentration of 150 pM and sequenced on an Illumina NovaSeq X Plus platform using 2 × 150-bp paired-end reads. The remaining libraries were normalized to 10 pM and sequenced on an Illumina MiSeq platform using 2 × 300-bp paired-end reads. All pooled libraries were denatured with freshly prepared 0.2 N sodium hydroxide before sequencing. These details matter because they document how a national public health laboratory, sequencing in-country, produced near-complete viral genomes during an active outbreak.

The bioinformatic pipeline began with quality control and trimming of raw paired-end reads using fastp v0.23.4, removing low-quality bases (Phred below 20), adaptor sequences, poly-G/poly-X tails, low-complexity regions and reads shorter than 50 base pairs. Host-derived reads were removed using HOSTILE v2.0.2, which aligns reads against the human reference genome GRCh38.p13; only read pairs that did not align to the human genome were retained. Non-host reads were then mapped to the Marburg virus reference genome (accession NC_001608.3) using BWA-MEM v0.7.18 in short-read mode with default Illumina-optimized parameters. The reference is the complete genome of the Musoke strain, derived from the 1980 outbreak in Kenya, a 19,112-base-pair sequence first sequenced in 1992. BAM files were sorted and indexed with SAMtools v1.17, and coverage, alignment quality and duplication rates were assessed with Qualimap v2.2.2.

Variants were called with iVar v1.3.1 combined with SAMtools mpileup, applying a minimum base quality of 20 and a variant frequency threshold of 0.03, generating per-sample consensus genomes with ambiguous positions coded as N. Consensus sequences were aligned with MAFFT v7.520 (L-INS-i). The final comparative dataset included 110 Marburg virus whole-genome sequences sampled between 1967 and 2025, spanning 13 countries: the DRC (n = 30), Rwanda (n = 18), Uganda (n = 17), Angola (n = 16), Ethiopia (n = 7), Sierra Leone (n = 6), Tanzania (n = 5), Kenya (n = 3), South Africa (n = 2), Ghana (n = 2), Guinea (n = 2), Germany (n = 1) and the Netherlands (n = 1). Clade assignments and quality control were performed with Nextclade v2.15.2, and phylogenies were reconstructed with IQ-TREE v2.2.5 under the best-fit substitution model identified by ModelFinder, with ultrafast bootstrap approximation for branch support.

To assess whether the data carried a usable temporal signal, the team performed root-to-tip regression analysis with TempEst v1.5.3 on the maximum-likelihood phylogeny, optimizing the tree root to maximize the correlation between root-to-tip genetic distance and sampling date, using RAVV as a true outgroup that was pruned before the regression and restricting the analysis to global Marburg virus genomes, following a previously validated approach. Time-scaled phylogenies were inferred with BEAST v10.5.0 under the HKY nucleotide substitution model with empirical base frequencies, selected by Bayesian information criterion comparison in ModelFinder. A fixed local clock model was used: a fixed substitution rate of 1 × 10−3 substitutions per site per year was assumed for the defined outbreak clade, representing a plausible evolutionary rate for an RNA virus during the early stage of an outbreak, while the deeper evolutionary history was allowed to evolve under a lognormal relaxed clock prior centred on 5 × 10−4 substitutions per site per year, reflecting the lower rate expected outside the rapid outbreak phase.

The demographic model was similarly two-phased. A coalescent Bayesian SkyGrid tree prior allowed effective population size to vary flexibly through time; the subtree from the Ethiopian lineage tMRCA to the present was modelled with an exponential growth coalescent prior, reflecting the rapid expansion expected during an outbreak, while the earlier, deeper history was modelled with a constant-population-size coalescent prior, assuming stable viral population dynamics before the emergence of the outbreak lineage. Two independent MCMC chains were run for 100 million states each, sampling every 10,000 steps, with the initial 10% discarded as burn-in. Convergence was evaluated visually in Tracer v1.7.2 and quantitatively through effective sample size calculations, with ESS values above 200 considered indicative of adequate mixing; all estimated parameters achieved ESS values above the recommended threshold. The remaining posterior samples, 45,001 per chain, were combined with LogCombiner v10.5.0, and tMRCAs were calculated as posterior mean node heights with TreeAnnotator v10.5.0, converted to calendar years using the decimal year of the most recent tip in the dataset.

Beyond the phylogeny, the researchers identified specific lineage-defining SNPs in the Ethiopian samples from multiple sequence alignments, mapped them to genes using reference GenBank annotations, translated them into amino acid changes and classified them as synonymous or non-synonymous. They also calculated nucleotide diversity, Watterson’s theta and segregating sites for the Ethiopian sequences, and estimated gene-level dN/dS ratios. The genome sequences generated in the study are publicly available on Pathoplexus under the accession numbers PP_00731R3.4, PP_00730R4.4, PP_00731Q5.4, PP_00731P7.4, PP_00730Q6.4, PP_0072YT4.4 and PP_0072ZR7.4, and all analysis code and data are available via GitHub. Taken together, the seven near-complete genomes, their placement in the MARV.A.1 clade and their high mutual nucleotide identity provide a genomic picture of the earliest phase of Ethiopia’s first recorded Marburg virus disease outbreak: a zoonotic spillover followed by human-to-human transmission, documented by genomic surveillance conducted within the affected country.

Subject of Research: Genomic surveillance of the first Marburg virus disease outbreak in Ethiopia

Article Title: Genomic analysis of early case samples from the 2025 Marburg virus outbreak in Ethiopia

Article References: Feyssa, M. D., Tollera, G., Abdella, S., Bashea, C., Zerihun, B., Abte, M., Getu, M., Agunie, A., Shure, W., Gebremicael, G., Kidane, E., Gonta, M., Weldemariam, A. G., Tesfaye, N., Kayema, N., Megersa, T. S., Geleta, D. A., Rameto, M. A., Kugna, F., … Wolday, D. (2026). Genomic analysis of early case samples from the 2025 Marburg virus outbreak in Ethiopia. Nature Microbiology. https://doi.org/10.1038/s41564-026-02540-y

Image Credits: AI Generated

DOI: 10.1038/s41564-026-02540-y

Keywords: Marburg virus, Ethiopia, genomic surveillance, outbreak, phylogenetics, zoonotic spillover, MARV.A.1 clade, whole-genome sequencing, viral haemorrhagic fever, case fatality rate, public health, epidemiology