GI-16 lineage (624/I or Q1), there and back again: The history of one of the major threats for poultry farming of our era

Franzo, Giovanni; Cecchinato, Mattia; Tosi, Giovanni; Fiorentini, Laura; Faccin, Francesca; Tucciarone, Claudia Maria; Trogu, Tiziana; Barbieri, Ilaria; Massi, Paola; Moreno, Ana

Since there was a clear bias in the number of sequences collected in each country and in different time periods, the original dataset was split in ten randomly generated databases in order to diminish the impact of this bias and evaluate the robustness of achieved results over several independent runs [32] . More specifically, a maximum of 5 sequences for each country-year pair were randomly selected and included in the study. For each dataset the time to Most Recent Common Ancestor (tMRCA), the evolutionary rates (substitutions/sites/year) and viral population dynamics (expressed as Effective population size (Ne)�generation time (t), also called relative genetic diversity) were jointly estimated in a Bayesian fashion using the serial coalescent approach implemented in BEAST 1.8.4 [33] . Additionally, the discrete state phylogeographic approach implemented in the same program was used to reconstruct the viral spreading pattern, assuming each collection country as a strain feature [34] . This method allows to estimate the location and time of ancestral strains, calibrated using the implemented molecular clock and population model. Additionally, the use of the Bayesian Stochastic Search Variable Selection (BSSVS) allows for a Bayesian Factor (BF) test that identifies the most parsimonious description of the spreading process, highlighting the most relevant and significant viral migration routes [35] . The nucleotide substitution model (i.e. GTR+Γ 4 ) was selected based on the results of Bayesian Information Criterion (BIC) calculated using Jmodeltest [28] , while molecular clock (relaxed lognormal molecular clock [35] ), population model (Bayesian skyline [36] ) and phylogeographic substitution model (symmetric substitution model) were selected based on the BF value, calculated using the marginal likelihood of different models, estimated with the Path Sampling (PS) and Stepping Stone (SS) method [37] .