Funded from 2009–2014, the Viral Evolution project studied the evolutionary dynamics of the influenza virus and other rapidly evolving pathogens. The team studied evolution both intra-host (within a single host) and inter-host (population level dynamics), as well as the way in which the former affects the latter.
All RNA viruses have roughly the same underlying mutation rate (~3 x 10^-5 errors per replication per base). Some of these viruses, such as measles, appear to be trapped in a genetic or antigenic “corner”, so that a one-time exposure to the virus results in near-lifetime immunity. Others, including the influenza virus, exhibit great genomic variability with sufficient fitness to allow them to evade immune and drug pressure, frequently jump across species barriers, and undergo significant neutral drift. The high mutation rate of influenza, coupled with its enormous antigenic and genomic plasticity, results in an infectious agent with the ability to respond in a significant manner to environmental, seasonal, behavioral and policy factors, well within the time-span of an epidemic or pandemic. Newly emerging viruses may fall anywhere in between these two extremes.
Given previous pandemics and panzoonotics, and the potential for emerging viruses to adversely affect the health of human and other animal populations, it is clearly essential to understand the factors that allow viruses to enter and spread through new host populations, and to rapidly adapt to their environment. It may ultimately be possible to predict what type of virus may emerge in populations in the future, where such emergence is likely to occur, and what species are most likely to act as reservoirs.
Dr. Rosenfeld is now on the Modeling & Forecasting team.
At the intra-host level, the team has developed an equation-based model (EBM) to study viral evolution, calibrated to specific upper and lower respiratory symptoms, which was then coupled to the population-level agent-based model (ABM; see information on FRED, below) to study intra- and inter-host evolution in a unified simulation framework.
At the inter-host level, the team is actively involved in the development of FRED, an agent-based modeling (ABM) platform being developed by a larger group of MIDAS researchers. Contributions of the evolution team to FRED include the development of:
- A multi-strain version of FRED, where any number of strains of a pathogen are propagated simultaneously in the population, each with its own phenotypes, such as transmissibility, antigenicity, drug resistance and virulence. Super-infection (simultaneous infection by more than one strain) and crossover or reassortment events can also be modeled.
- A genotype-phenotype mapping representation in FRED, where strains can differ in their transmissibility, antigenicity, drug-resistance and virulence.
- A generalized immunity mechanisms which can represent both strain-transcending and strain-specific immunity, based on the retained exposure history of each agent.
- A “middle layer” to interface the intra-host equation-based model (EBM) to the population-level agent based model (ABM).
The team then used FRED to examine research topics of interest, including:
Antiviral Drug Policies
Using FRED, Viral Evolution team members studied the impact of alternative antiviral drug policies on epidemic size and emergence of drug resistance. They found that early anti-viral drug treatment results in significant variability in the prevalence of drug resistance, suggesting that, under some conditions, aggressive antiviral drug intervention should be slightly delayed to minimize the risk of widespread resistance.
Infectivity and Symptomaticity
Using explicit models of infectivity (viral load trajectory) and symptomaticity (immune response trajectory) informed by the intra-host model, the team studied the impact of alternative infectivity and symptomaticity dynamics on key epidemic outcomes. They found that in the realistic agent population structure of FRED, average infectivity and average generation time are insufficient for predicting epidemic outcomes, suggesting a need for more detailed disease surveillance.
Using ultra-deep sequence analysis of data obtained from an influenza A/H1N1 epidemic, the team found the presence of drug-resistant virus before drug treatment in an immune-comprised child, as well as evidence for transmission of drug-resistant virus in a household. Elodie Ghedin, Edward C Holmes, Jay V DePasse, Lady Tatiana Pinilla, Adam Fitch, Marie-Eve Hamelin, Jesse Papenburg and Guy Boivin. Presence of oseltamivir-resistant pandemic A/H1N1 minor variants before drug therapy with subsequent selection and transmission. The Journal of Infectious diseases 206(10):1504–11, November 2012. PMCID: PMC3475640
Large-scale population models of influenza A epidemics assume limited variability in strain phenotype as they infect targets, and may better form a finer description of host variability and virus phenotypic variations. Host-levels models offering such a phenotypic description are linked to population models in a computationally efficient manner to describe this variability. Sarah Lukens, Jay V DePasse, David Swigon, Danielle Miller, Roni Rosenfeld, Elodie Ghedin and Gilles Clermont. Linking population-level and host-level models of influenza A virus. Journal of Critical Care 27(3):e12, June 2012.
A small ODE model of intrahost immune response to bacterial pneumonia can explain multiple phenotypes in murine pneumonia data. The model demonstrates the importance of a strong immune response both by phagocytosis in blood and by the innate immunity (strong nonspecific defenses are not sufficient to overcome infection, and extrapulmonary phagocytosis is key to survival of pneumonia). We have also offered a model-based method of contrasting phenotypic responses to infectious challenges, potentially broadly applicable in heterogeneous populations and experimental challenges. Mochan E, Swigon D, Ermentrout GB, Lukens S, Clermont G. A mathematical model of intrahost pneumococcal pneumonia infection dynamics in murine strains. J Theor Biol. 2014 Mar 2;353C:44-54. doi: 10.1016/j.jtbi.2014.02.021.
A Probabilistic Finite State approximation of the early stage of an Epidemic
Agent Based Models are general and flexible, allowing for arbitrary levels of detail and realism in modeling epidemics. However, they are much harder to draw general conclusions from, relative to the more analytical compartment based models. Furthermore, while agent-based models can in principle capture the stochastic nature of epidemics, they require multiple realizations to empirically estimate it.
In the early stages of an epidemic, its dynamics is particularly variable and hence important to capture. For example, Influenza strains imported early in the season may or may not come to dominate that season, depending on small variations in the importation schedule and other chance effects. To address this problem, the evolution team developed a probabilistic finite state approximation of the early stage of an epidemic, which can be fitted to any particular city and any ABM setting. This allows them to calculate analytically the distribution of epidemic trajectories for any desired seeding schedule without having to run many realizations, and is an important step towards global modeling of seasonal antigenic variation.