We extracted total genomic DNA from silica-dried leaf tissue using Qiagen's DNeasy Plant Mini Kit. All DNA extracts were quantified on a Nanodrop 1000 system (Thermo Fisher Scientific). We combined DNA extracts for 24 individuals from each species and sampling locality into DNA pools of equal molarity. All DNA pools were individually barcoded and sequenced on both the SOLiD5500 and Illumina HiSeq systems at the Functional Genomics Center Zurich, using 75 and 100bp paired end chemistry, respectively. High-quality reads were reference-mapped against the P. trichocarpa v3 / build 210 genome sequence, using allelic information available for P. alba and P. tremula to generate majority-rule consensus reference sequences. We formatted alignments with Picardtools 1.125 and realigned around indels using GATK version 3.3.0. We tabulated the depth of coverage for any given variable site using DepthOfCoverage in GATK, filtering reads at a minimum base quality of 20 and a minimum mapping quality of 20. We only retained sites with a minimum coverage of eight reads in each pool, and a minimum coverage of three reads for the minor allele to avoid spurious SNP calls. We used the fractions of raw sequencing coverage for each allele and site as estimates of population allele frequencies. Filtering for demographic inference from SFS: We filtered the pool-seq data using the 1st and 3rd quartile of the depth distribution. This was done in order to account for variation in coverage depth between populations and remove loci in the tails of the distribution, likely representing sequencing errors in the casec of low coverage or duplicated loci in the case of high coverage. We thinned the pool-seq data for the Italian locality to randomly keep one SNP per window of 8000 base pairs (bp). The joint SFS for P. alba and P. tremula in the Italian locality was estimated from the data with the program ∂a∂i (Guntenkunst et al. 2011) using P. trichocarpa as an outgroup to distinguish ancestral and derived alleles. To keep the maximum number of SNPs that matched our stringent coverage thresholds (above), the SFS was downsampled o 20 reads (assumed to correspond to 10 diploid individuals) for each population