Schematic diagram of the method for identifying GCNVs likely under parallel selection. (A) Re-sequenced reads (thin lines) from each individual were mapped to the stickleback reference genome (thick lines). (B) The numbers of mapped reads that overlapped with genes were counted, and we searched for genes that showed significant differences in the normalized read numbers between the freshwater (closed circles) and marine groups (open circles) with a false discovery rate (FDR) < 0.05. Genes that showed significant differences under the three mapping options were regarded as GCNVs likely under parallel selection. (C) The number of different allelic sequences was counted for each of the identified GCNVs by enumerating every pair of SNV positions that was located within the read length. If three or more allelic sequences were observed for a gene, the GCNV involved duplications or multiplications.