Adaptation to natural facial categories

We assembled two independent data sets (ferns and angiosperms) and used the lycophyte Huperzia as the outgroup (not shown in Fig. 1) for both. For ferns, we sampled the chloroplast rbcL and rps4 genes for 45 taxa: 41 leptosporangiate ferns from all major lineages (focusing within the polypods), one horsetail, two seed plants, and the outgroup. For angiosperms, we sampled the chloroplast rbcL, atpB and nuclear small-subunit ribosomal DNA genes for 95 taxa (mostly a subset from data set in ref. 14): 84 angiosperms, eight gymnosperms, one leptosporangiate fern, one horsetail, and the outgroup. Most DNA sequence data were already available in GenBank; 19 new fern sequences were generated as part of this study following the DNA extraction, amplification and sequencing protocols in ref. 18, and are deposited in GenBank. For fern voucher information and GenBank accession numbers see Supplementary Information. Phylogenetic analyses using a bayesian approach were conducted separately for ferns and angiosperms with MrBayes version 3.0b4 (ref. 29). Each gene in each data set was assigned its own model of evolution (GTR þ I þ G for each gene, determined using a hierarchical likelihood ratio test approach) and analyses were conducted using four chains, run for a total of 10,000,000 generations, with trees sampled from the cold chain every 1,000 generations. The resulting 10,000 trees were plotted against their likelihoods to determine the point where the likelihoods converged on a maximum value, and all 500 trees (500,000 generations) before this convergence were discarded as the 'burn-in' phase. For each data set (ferns and angiosperms), we computed a majority-rule consensus of the remaining 9,500 trees. We used this phylogenetic hypothesis (with average branch lengths), as well as 1,000 trees randomly sampled from among the 9,500 trees, in the subsequent analyses. Divergence time estimates were obtained through penalized likelihood analyses (truncated Newton algorithm) of the fern and angiosperm consensus phylogenies in r8s version 1.60 (ref. 30). For each data set, fossil age constraints were applied as indicated in Table 1 (for details see Supplementary Information) and the appropriate smoothing value was determined using cross-validation. For angiosperms, fossil age constraints 22 (Table 1) were applied in two ways, differing only in the application of the fossil age of the angiosperm crown group (node b10 in Fig. 1b). The first approach was strict (age fixed at 132 Myr; ref. 22) and the second relaxed (minimum age constraint of 132 Myr applied). To …