CHAPTER 15 Dealing with model uncertainty in reconstructing ancestral proteins in the laboratory: examples from archosaur visual pigments and coral fluorescent proteins Belinda S.W. Chang, Mikhail V. Matz, Steven F. Field, Johannes Mu ¨ ller, and Ilke van Hazel 15.1 Introduction Resurrecting ancestral proteins in the laboratory can be a powerful tool in studies of protein structure and function as they can offer a rare glimpse into the evolutionary history of molecular function (Malcolm et al., 1990; Adey et al., 1994; Chandrasekharan et al., 1996; Dean and Golding, 1997; Bishop et al., 2000; Chang and Donoghue, 2000; Sun et al., 2002; Zhang and Rosenberg, 2002; Thornton, 2004). Another, perhaps even more intriguing reason for reconstructing ancestral pro- teins lies in the hope of achieving a better under- standing of the biology of ancient animals that may have possessed these proteins (Jermann et al., 1995; Messier and Stewart, 1997; Galtier et al., 1999; Benner, 2002; Gaucher et al., 2003). Proteins that are involved in sensory systems might be parti- cularly revealing with respect to the physiology and behavior of ancient animals that can no longer be studied directly in the laboratory (Nei et al., 1997; Boissinot et al., 1998; Chang et al., 2002). Moreover, experimental tests of laboratory-recre- ated ancestral proteins would provide information different from that obtained through studies of fossils. Although interpretations based on recrea- tions of single molecules are of course limited, under the best of circumstances one may hope to test some of the theories of ancient animal biology derived from other methods such as paleontolo- gical studies (Chang et al., 2002). Reconstructions of the past depend entirely on the accuracies and limitations of the statistical methods and models employed. However, even in cases of deep divergences, when the accuracy of the reconstruction may be low, the experimental outcome remains valuable, as the effects of altering specific amino acids on a protein’s structure and function can be interesting, independently of whether or not the sequence represents the true ancestor. The general approach of using site-directed mutagenesis methods to alter amino acids in order to assess shifts in function is one of the most widely employed in studying protein function. Advances in phylogenetic methods of ancestral reconstruction, particularly the development of likelihood/Bayesian models that incorporate many different aspects of sequence evolution, have led to a plethora of models and methods available for use in phylogenetic reconstruction in recent years (Thorne, 2000; Huelsenbeck and Bollback, 2001; Whelan et al., 2001; Nielsen, 2005). This chapter briefly discusses some of the models available for use in ancestral reconstruction, then describes ways to address variation in reconstructed ancestral sequences when the intent is to recreate proteins 164