We used the program STRUCTURE v2.3.4 [16]

NA extractions were attempted for all taxa for which recent and sufficient tissues were available. Of the samples that were successfully extracted, at least two individuals for each taxon were included when available. This yielded 44 Malagasy Canarium accessions, to which we added an additional four Southeast Asian Canarium species to serve as outgroups. Genomic libraries were prepared for genotyping-by-sequencing (GBS) following the protocol described by Escudero et al. [14], but with the addition of a size selection step. After restriction digestion, fragment sizes were visualized on an Agilent 2100 Bioanalyzer and selected in the range of 300–800 bp using a Pippin Prep system. The final library containing 48 barcoded individuals was sequenced on two lanes of an Illumina HiSeq 2500 at Yale University’s Center for Genome Analysis to generate 75 bp single-end reads.

Fully reproducible code is available in the form of jupyter notebooks (https://jupyter.org) with instructions to install all necessary software, download sequence data, assemble it, and run genomic analyses (https://github.com/dereneaton/Canarium-GBS) (DOI 10.5281/zenodo.1273357). For all genomic analyses in this study we used the ipyrad Python API and the ipyrad-analysis toolkit (https://github.com/dereneaton/ipyrad) which provides a set of wrappers around common genomic analysis tools to run highly parallelized analyses using simple Python scripts.