We used the deep sequenced brain samples to estimate the proportion of sampling artifacts in terminal branches, and effectively subtracted the proportion of artifacts to obtain reliable phylogenetic signals. Each brain sample was split in three completely independent samples of 100 million reads. Top: Trees constructed using: regions covered only with one read in each taxon, regions covered by 1 and 5 reads (very low expression), regions covered by any reads, regions above 10 reads (mid expression) and regions above 100 reads (high expression). The percentage shown indicates the average level of sampling artifacts for each threshold, derived from the length of the terminal branches not found in all replicates of each taxon, i.e. the uncorrelated portion across samples of the same origin. These numbers are highest for the lowly expressed regions, and are lowest for the highly expressed regions, and are more or less constant within comparisons. Once subtracted, the phylogenetic signal remains robust. Taxon names as summarized in Figure 1. The figure part with the 1 read fraction corresponds to Figure 4C.