Genomic signatures of host resistance to virus infection in birds

What makes an individual or species Resistant, Tolerant, or Susceptible (R/T/S) to a virus? Deep epidemic knowledge for well-studied species, empowered by the torrent of genome data generated for birds1, together with advances in Deep Learning2 allow us to identify predictors of R/T/S in genomic, ecological and environmental data. Datasets with detailed spatiotemporal R/T/S data to train these Deep Learning models are rare; this is where our project stands out. Our team comprises evolutionary geneticists, bioinformaticians, virologists, AI modellers and conservation biologists with a track record in host-pathogen systems and avian genomics. We have gathered data on viral outbreaks for 18 years and sequenced over 200 bird genomes, working on some of the best-studied avian models. Here, we will harness data from parrot-circovirus host-pathogen systems to develop an AI model-informed approach to identify environmental, ecological and genomic factors that can predict whether individuals or populations are Resistant, Tolerant or Susceptible to a particular pathogen. We will use this information to better predict long-term viability and extinction risk of threatened parrot species. Population size decline leads to loss of immunogenetic variation, which makes threatened species particularly susceptible to (re)emerging infectious diseases. Furthermore, identifying which wildlife species are most at risk of becoming reservoir hosts will help us predict3 spill-over events across species4. Circoviruses are highly prevalent in bird species, including pigeons (pigeon circovirus), parrots5 (psittacine beak and feather disease virus; BFDV), and chickens (chicken anaemia agent virus). In fact, a total of 110 bird species are known to be infected by circoviruses, increasing the extinction risk of some of the most threatened species. Across species, spillovers extend this risk to other species, as circoviruses also infect plants, cattle, bats and humans6. One of the commonest circoviruses in birds is Beak and Feather Disease Virus (BFDV), a globally widespread ‘generalist’ circovirus with a compact circular, single-stranded 2kb DNA genome. A close phylogenetic association exists between BFDV and its psittacine hosts, suggesting coevolution7. Amongst birds, parrots are highly phylogenetically conserved8, providing an ideal system for predicting R/T/S individuals and across species. We will analyse an intensively studied natural population of the Mauritius parakeet which has experienced an exceptionally well-documented outbreak of BFDV. Individual infection histories of clinical symptoms are available for >1,200 individuals over 28 years, and a blood-sample archive of >2,700 individuals. First, we will examine 400 transcriptomes and 1500 genomes of birds with known R/T/S from our archive. We will use genomic metrics paired with ecological, epidemic and fitness data to train a Deep Learning model to identify predictors for the outcome of an infection in our model species. We will then test our model on populations of three additional species with known R/T/S-status which have also experienced well-characterised outbreaks (ringneck parakeet, orange-bellied parrot, and red-crowned parakeet). Next, to test if our trained model can identify risk in less well-characterised species, we will critically test its predictive power on genomes of 14 matched ‘species-pairs’ (total 28 parrot genomes). Each comprises a putatively ‘susceptible’ (S) and a ‘resistant’ (R) or tolerant (T) species based on published data on prevalence and clinical symptoms of BFDV. Finally, we will evaluate how our predictive model performs by testing for presence of BFDV in host individuals in all parrot species with genomic data present in European zoos (total zoo population of 18,500 individuals from >300 species).