with Model A and Model A1 (null model). Both models require a priori specification of the lineages likely to have experienced positive selection. Visual inspection of Fig. 10.3 allowed us to test two blue-shifted lineages that may have experienced positive selection. Sites that may be under positive selection in the spectrally shifted lineages were identified with a Bayes Empirical Bayes approach (Yang et al., 2005).

We used homology modeling to study the relationship between the L. arthemis astyanax opsin structure and function using the methods described (Briscoe, 2002). The full-length L. arthemis astyanax L opsin protein sequence was manually aligned with the bovine template 1U19.pdb (Okada et al., 2004), and the alignment was submitted to the Swiss-Model server (www.expasy.ch/swissmod) (Schwede et al., 2003). The atomic coordinates were then viewed with SwissPdb Viewer (www.expasy.ch/spdbv) (Guex and Peitsch, 1997), and candidate spectral tuning sites were mapped onto the 3D homology model.

We thank A. P. Platt (University of Maryland, Baltimore County, Baltimore, MD), A. Porter (University of Massachusetts, Amherst, MA), and C. L. Remington (Yale University, New Haven, CT) for providing specimens; C. Wang and W. Huynh for technical assistance; and A. Kelber, D. Osorio, N. Pohl, M. Brandt, N. Aguilar-Roca, and K. Thornton for useful comments on earlier versions of this manuscript. We especially thank S. M. Reppert for insightful discussions. This work was supported by National Science Foundation Grant IOB-0346765 (to A.D.B.)