Summary of PAML discrete ω ratio variation models. Each point on the plots in (A), (B), and (C) corresponds to the frequency and ω ratios for the best minimum model (e.g., M0, M3/ncatG = 2, M3/ncatG = 3, etc.) that provides a statistically superior fit to the data (i.e., a more complex model with additional codon ω ratio classes that does not provide a statistically better fit to the next simplest model is rejected). For example, the analysis of the full-length sequence of NR1A1 (TRα) for all available vertebrate species shows that M3/ncatG = 3 is superior to M3/ncatG = 2 but statistically equivalent to M3/ncatG = 4. Consequently, plotted on Figure 1 are three points for the NR1A1 M3/ncatG = 3 analysis corresponding to frequency and ω ratios for three classes of codons – 80.6% (frequency = 0.806) of codons have an estimated ω ratio of 0.004, 14.9% have an ω ratio of 0.094, and 4.5% have an ω ratio of 0.259. An analysis that shows M0 is the best minimum model will have 100% of codons (frequency = 1.0) with a particular ω ratio. (A), (B), and (C) apply to analyses of full-length sequences, DBD only, and LBD only, respectively. The open circles are for analyses of all available vertebrate sequences while the closed circles are for analyses of mammals only. For part (C), the red open and closed triangles represent data for the LBD of the AHR gene (a non-NR gene that encodes a protein with similar function to PXR and CAR).