Although several filoviruses are dangerous human pathogens, there is conflicting evidence regarding their origins and interactions with animal hosts. Here we attempt to improve this understanding using the paleoviral record over a geological time scale, protein structure predictions, tests for evolutionary maintenance, and phylogenetic methods that alleviate sources of bias and error. We found evidence for long branch attraction bias in the L gene tree for filoviruses, and that using codon-specific models and protein structural comparisons of paleoviruses ameliorated conflict and bias. We found evidence for four ancient filoviral groups, each with extant viruses and paleoviruses with open reading frames. Furthermore, we found evidence of repeated transfers of filovirus-like elements to mouse-like rodents. A filovirus-like nucleoprotein ortholog with an open reading frame was detected in three subfamilies of spalacid rodents (present since the Miocene). These elements were unique among the detected filovirus-like paleoviruses in possessing open reading frames, expression products, and evidence for purifying selection. Our finding of structural conservation over geological time for paleoviruses informs virus and paleovirus discovery methods. Our results resolve a deep conflict in the evolutionary framework for filoviruses and reveal that genomic transfers to vertebrate hosts with potentially functional co-options have been more widespread than previously appreciated.Author SummaryFiloviruses are a family of RNA viruses discovered in 1967 and notorious for spillover of the dangerous pathogens, Ebola virus and Marburg virus. However, their origins, deeper relations, diversity, and interactions with animal hosts remain controversial. Part of the confusion may be that differing rates of evolution among divergent viral lineages can create a bias termed long branch attraction (LBA). We tested for this scenario in the L protein gene sequence of filoviruses and found evidence that LBA is occurring leading to a false pairing of filovirus lineages associated with a fish and a snake. We found that using nucleotides instead of amino acids when inferring trees, paleoviral sequences with open reading frames, additional conserved genes, and comparisons of predicted protein structures can resolve the LBA. We found four major groups of filoviruses, with the paleoviral record and trees being consistent with a fish origin for the family. Moreover, we found evidence of a filovirus-like element in spalacid rodents that has been evolutionarily maintained at the open-reading frame, amino acid sequence and structural level for over 20 million years. This element was also expressed in the liver, a target of filoviral infections. We conclude that genomic interactions of filoviruses with vertebrates, including the co-option of viral genes, are more important than previously appreciated.
