Stochastic mapping of geographical trait for phylogeography inference

The reference-based maximum likelihood phylogeny of 395 CenAsiaIII S. sonnei served as an input for reconstructing the ancestral geographical states of each isolate, as adopted from a recently described approach. We treated geographical source (sub-continental level) of the organisms as discrete characters. Due to biased sampling toward South Asia, we sub-sampled the phylogeny to include equal numbers of isolates from each character state (i.e. geographical region; n=13 from South Asia, Africa, Europe, and combined Southeast Asia/East Asia), generating 1,000 subsampled trees. Regions with fewer than 13 isolates (i.e. Middle East and America) were excluded. We utilized stochastic mapping to quantify transition events between geographical characters as well as the total time spent within each character state, separately for each subsampled tree. Stochastic mapping, implemented as the function make.simmap in the R package phytools v0.6.0, was performed under an asymmetric model of character change (ARD) with the rate matrix sampled from the posterior probability distribution using MCMC (Q=mcmc) for 100 simulations. To ensure that resultant phylogeographical signal was not an artefact due to over-sampling in South Asia, we permuted (without replacement) the tip-location of the original maximum likelihood phylogeny, generating ten independent randomization sets. We repeated stochastic mapping analysis for each randomization as detailed above, using only 500 subsampled trees from each randomization to reduce computational expense. Results from the ‘true’ and ‘randomization’ runs were compared by analysis of variance (ANOVA) with post-hoc Tukey test.