Research
We study the epidemiology and evolution of a broad range of pathogens using a mix of computational and experimental methods. On the computational side, we develop methods for genomic epidemiology and phylodynamics. By relating the phylogenetic history of pathogens to their population dynamics, these phylodynamic methods allow us to reconstruct the spread of pathogens through space and time from genomic sequence data. Recent work has focused on extending phylodynamic methods to consider how multiple selection pressures and recombination shape pathogen phylogenies. On the experimental side, we study how RNA viruses adapt to novel hosts and (potentially) overcome fitness trade-offs between hosts to expand their host range.
We apply our phylodynamic methods to gain insights into the epidemiology and evolution of a wide range of pathogens, including both human and agricultural pathogens. We are particularly interested in generalist plant viruses with broad host ranges, and use tomato spotted wilt virus (TSWV) as a model system to explore the evolution of viral host range in the face of fitness trade-offs between hosts.
Advancing phylodynamic methods for genomic epidemiology
Phylodynamics for rapidly adapting pathogens
Microbial pathogens rapidly adapt to novel selection pressures, allowing them to infect new hosts, escape host defenses and evolve resistance to antimicrobials. Yet almost all current phylodynamic methods assume sequence evolution is neutral such that mutations cannot feedback and impact pathogen fitness. Because this assumption is obviously unrealistic, we are extending phylodynamic birth-death models to consider non-neutral or adaptive evolution where mutations and other genetic variation can influence pathogen fitness.
These non-neutral phylodynamic models can be used to gain a better understanding of what factors shape pathogen fitness—and thereby epidemic potential—in real world settings. In particular, we have applied these methods to estimate the fitness effects of individual mutations, allowing us to identify the mutations driving pathogen adaptation (Rasmussen and Stadler, 2019; Kepler et al., 2021). We also recently applied these methods to estimate the fitness benefits (and costs) of antimicrobial resistance genes in bacterial pathogens like E. coli ST 131 (Kepler et al., 2024)
Recombination-aware phylodynamics
Recombination and the horizontal transfer of genetic material between lineages tremendously complicates the phylogenetic history of many pathogens. In the presence of recombination, the ancestral relationships among sampled individuals can no longer be captured by a single phylogenetic tree, but rather a mosaic of different ancestral relationships along the genome. Ancestral recombination graphs (ARGs) provide one way to capture this mosaic ancestry and we are adapting phylodynamic methods to allow for demographic inferences from ARGs (Guo et al., 2022). We also recently released Espalier, a Python package for working with trees in the presence of phylogenetic discordance caused by recombination and for rapidly reconstructing approximate ARGs (Rasmussen and Guo, 2023).
Yet working with ARGs for pathogens with high recombination rates remains challenging. Ongoing work is therefore exploring best practices for locating recombination breakpoints/events (Cen and Rasmussen, 2024) and novel methods for visualizing and simplifying ARGs.
Optimizing sampling in population genomics and genomic epidemiology
Ultimately, how much information genomic sequences provide about population history depends on how they are sampled. Yet surprisingly little attention has been given to optimizing genomic sampling in population genomics and genomic epidemiology. Recent work therefore developed a sequential decision making framework to optimize genomic sampling using Markov decision processes (Rasmussen et al., 2025). These MDPs allow us to efficiently compute the expected long-term value of a given sampling strategy and therefore structure and guide our search towards optimal strategies. Ongoing work builds off of these MDPs using reinforcement learning to identify optimal sampling strategies in more complex settings.
Fitness trade-offs and the evolution of viral host range
Pathogens frequently jump between host species and adapt to novel hosts. However, fitness trade-offs may constrain the range of hosts a given pathogen can infect if adapting to one host entails a fitness cost in another host. Such constraints raise the question of how generalist pathogens with broad host ranges seemingly overcome such trade-offs.
We use tomato spotted wilt virus (TSWV) to explore how fitness trade-offs shape viral host ranges. In addition to infecting agronomically important crops, TSWV has a remarkably broad range of host plants (>1000 described species). Past work therefore combined experimental evolution with deep sequencing to look at how such a “super-generalist” adapts to alternative hosts (Ruark-Seward et al., 2020). These experiments revealed that a surprisingly large fraction of genetic variants are beneficial in different plant hosts and in thrips, the virus’s insect vector.
Our work with TSWV has inspired further work into how viruses evolve in the face of fitness trade-offs. For example, theory has emphasized mutations often have opposite or antagonistic effects on fitness in different environments, leading to strong fitness trade-offs. Yet our results from TSWV suggest that mutations often have concordant fitness effects across environments (i.e. positive pleiotropy). A recent meta-analysis on the pleiotropic fitness effects of mutations in different hosts revealed that this is true in viruses more generally, suggesting that positive pleiotropy may soften trade-offs hosts due to the availability of mutations with fitness benifits across hosts (Wang et al., 2024).