We determined how a large and diverse group of viruses, those with circular single-stranded DNA genomes that encode a particular Rep protein to help replicate their genomes (CRESS DNA viruses), are related to each other. Using the only protein shared by all of these viruses (Rep) we built a family tree with over 900 species of CRESS DNA viruses. We determined that the evolution of Rep proteins in these viruses was not described very well by the standard tools that evolutionary biologists use to describe protein evolution in bacteria and eukaryotes. This motivated us to create a new model for Rep protein evolution. This bespoke model improved our ability to see deep relationships among families of CRESS DNA viruses compared to standard models. The most exciting conclusion from our tree (image 1) is that the Reps of families Geminiviridae (green) and Genomoviridae (pink) are not separate groups, but it is instead clear that genomovirus Rep evolved from a geminivirus Rep, specifically a Rep that has to be made in multiple parts and spliced together to function. This splice-form Rep appears to have evolved only one among these families, and is used by four genera of geminiviruses, almost all genomoviruses, and by the black unclassified Rep sequences that were in the same group on the tree. It is logical that the more complex splice-form Rep would have evolved once instead of multiple times in a similar way, but our study was the first time this mixed relationship among geminivirus and genomovirus Reps has been observed. The high confidence we have in the relationship is due to strong statistical support from analyses using our bespoke model for CRESS DNA virus evolution. We have made our model freely available and incorporated it into a common tool to determine which evolutionary model to use (ProtTest, screenshot in image 2) and this, and much more information about CRESS DNA viruses is available at cressdna.org.

We also used four types of machine learning to help automate the task of figuring out which CRESS DNA virus family newly sequenced genomes belong to and to find new patterns in CRESS DNA viruses that can help us understand their evolution. Surprisingly, a simple metric that was useful for classifying CRESS DNA viruses was amino acid content of the Rep (or also of another common protein in CRESS DNA viruses, the coat protein). This is a useful, fast and independent complement to current classification methods that all rely on comparing the order of amino acids (or DNA) between sequences. One of our machine learning approaches also identified features of each genus of CRESS DNA viruses that were useful for classification, and we plan to investigate whether these features have biological significance to these viral groups.

This project trained three graduate students, two postdocs and two undergraduates in computational biology skills, data analysis and scientific communication. Members of the lab participated in outreach activities at local middle schools, including teaching over one hundred 7^th graders about how protein evolution occurs.

Last Modified: 07/05/2019
Modified by: Siobain M Duffy