The ancestors of the archosaurs, a major branch of the diapsid reptiles, originated more than 240 MYA near the dawn of the Triassic Period. We used maximum likelihood phylogenetic ancestral reconstruction methods and explored different models of evolution for inferring the amino acid sequence of a putative ancestral archosaur visual pigment. Three different types of maximum likelihood models were used: nucleotide-based, amino acid–based, and codon-based models. Where possible, within each type of model, likelihood ratio tests were used to determine which model best fit the data. Ancestral reconstructions of the ancestral archosaur node using the best-fitting models of each type were found to be in agreement, except for three amino acid residues at which one reconstruction differed from the other two. To determine if these ancestral pigments would be functionally active, the corresponding genes were chemically synthesized and then expressed in a mammalian cell line in tissue culture. The expressed artificial genes were all found to bind to 11-cis-retinal to yield stable photoactive pigments with λmax values of about 508 nm, which is slightly redshifted relative to that of extant vertebrate pigments. The ancestral archosaur pigments also activated the retinal G protein transducin, as measured in a fluorescence assay. Our results show that ancestral genes from ancient organisms can be reconstructed de novo and tested for function using a combination of phylogenetic and biochemical methods.
Visual pigments trigger the critical first step in the biochemical cascade of vision (Stryer 1986<$REFLINK> ). Several key features of visual pigments have been conserved throughout evolution. For example, a retinylidene chromophore is linked covalently via a Schiff base to a highly conserved lysine residue in all cases. But visual pigments have also evolved to perform specialized functions, such as color vision and scotopic (dim light) vision. Rhodopsin is specialized for high sensitivity under conditions of dim light. It has evolved several unique photochemical and biochemical properties, including an unusually high quantum efficiency, an extremely low level of biochemical noise in darkness, and an absorption maximum at about 500 nm (Menon, Han, and Sakmar 2001<$REFLINK> ).
One experimental approach used in studies of molecular evolution is the use of phylogenetic methods to infer ancestral sequences of biological molecules with the aim of recreating extinct genes or proteins in the laboratory (Chang and Donoghue 2000<$REFLINK> ). This approach shows much promise for investigating the function and evolution of ancient proteins (Malcolm et al. 1990<$REFLINK> ; Adey et al. 1994<$REFLINK> ; Chandrasekharan et al. 1996<$REFLINK> ; Dean and Golding 1997<$REFLINK> ; Bishop, Dean, and Mitchell-Olds 2000<$REFLINK> ), and perhaps even the organisms in which they existed (Jermann et al. 1995<$REFLINK> ; Messier and Stewart 1997<$REFLINK> ; Nei, Zhang, and Yokoyama 1997<$REFLINK> ; Boissinot et al. 1998<$REFLINK> ; Galtier, Tourasse, and Gouy 1999<$REFLINK> ). But in taking these studies of ancestral proteins into the laboratory, few experimental studies have explored the use of maximum likelihood methods of ancestral reconstruction, particularly in the light of the plethora of likelihood models now available. Using maximum likelihood methods (Felsenstein 1981<$REFLINK> ; Yang, Kumar, and Nei 1995<$REFLINK> ) we explored different models for reconstructing an ancestral archosaur rhodopsin. Once inferred, the phylogenetically reconstructed archosaur rhodopsin gene sequences were then synthesized, expressed, and assayed for function in the laboratory. The ancestral archosaurs were chosen as a test case for this type of molecular paleontological approach for two reasons. First, although the archosaur lineage gave rise to some of the largest reptiles to walk the earth, including the late Cretaceous carnivorous dinosaurs, little is known yet of the physiology and behavior of their ancestors. Because visual pigments constitute the critical first step in the visual phototransduction cascade in the eye and rhodopsin in particular is essential for vision at low light levels, recreating the inferred visual pigments of the archosaur ancestors in the laboratory should be an important initial step toward a better understanding of their visual capabilities that is difficult to obtain using other means. Second, divergences among extant archosaur rhodopsin protein sequences are no more than 16%, levels within the range at which likelihood methods of ancestral reconstruction should work reasonably well.
Materials and Methods
Thirty vertebrate rhodopsin nucleotide sequences were obtained from GenBank, aligned with ClustalW, and adjusted by eye to ensure the alignment of structurally important amino acids and that no gaps existed within codons. Because the amino acids at the extreme ends of rhodopsin (21 amino acids at the N-terminus and 25 amino acids after the palmitoylation site at the C-terminus) are not thought to be important in determining any aspect of either photon absorption or transducin activation and are more problematic to align with confidence, they were excluded from the ancestral reconstruction analysis (with corresponding residues from bovine rhodopsin used for the synthetic archosaur gene). The aligned vertebrate rhodopsin sequences were used to reconstruct the ancestral archosaur sequence using the phylogeny in figure 1 , which reflects current understanding of systematic relationships among the major vertebrate lineages (Carroll 1997<$REFLINK> ; de Jong 1998<$REFLINK> ).
Ancestral reconstructions of the archosaur node were performed using maximum likelihood methods (Felsenstein 1981<$REFLINK> ; Yang, Kumar, and Nei 1995<$REFLINK> ), as implemented in the PAML program (Yang 1997<$REFLINK> ). Marginal posterior probabilities at each site were also calculated in this program. It is important to note that these posterior probabilities are computed using the empirical Bayes approach, which does not account for errors in the parameter estimates or in the phylogeny itself. Where possible, pairwise likelihood ratio tests were used to assess among nested models which best fit the data at hand, a common approach for identifying models for use in likelihood analyses (Navidi, Churchill, and von Haeseler 1991<$REFLINK> ; Yang, Goldman, and Friday 1994<$REFLINK> ; Sullivan and Swofford 1997<$REFLINK> ).
Synthetic Gene Design, Construction, and Expression
The artificial archosaur gene was synthesized in large fragments (168 to 230-mers) on a solid-phase oligonucleotide synthesizer (Applied Biosystems, model 392). The synthesized fragments were amplified using the Pfu polymerase (Stratagene), cloned into the pCR-Blunt vector (Invitrogen), pieced together using unique restriction sites, and then cloned into a mammalian expression vector (pMT). The artificial archosaur gene was expressed and purified by previously described methods (Han et al. 1996<$REFLINK> ), essentially by transient transfection into COS cells using Lipofectamine Plus (Life Technologies), harvested after 48 h, regenerated in 5 μM 11-cis retinal, solubilized in 1% n-dodecyl-β-d-maltoside detergent, and immunoaffinity purified using the 1D4 monoclonal antibody. Absorbance spectroscopy was performed at 25°C using a Perkin-Elmer Lambda 800 spectrophotometer, using quartz cuvettes with a 1-cm pathlength. Transducin fluorescence was monitored at 10°C using an SPEX spectrofluorometer equipped with a Xenon arc lamp by methods described previously (Marin et al. 2000<$REFLINK> ).
Results and Discussion
Reconstructions of the ancestral archosaur rhodopsin were performed using the data set of vertebrate rhodopsin genes available in GenBank and a phylogeny (fig. 1 ) that reflects current understanding of systematic relationships among the major vertebrate lineages (Carroll 1997<$REFLINK> ; de Jong 1998<$REFLINK> ). Phylogenetic reconstructions using model-based methods such as maximum likelihood are known to be sensitive to the assumptions of the particular model used in the analysis. Oversimplified models can yield misleading phylogenetic reconstructions (Cao et al. 1994<$REFLINK> ; Huelsenbeck 1997<$REFLINK> ) and may also be problematic when reconstructing ancestral states (Chang and Donoghue 2000<$REFLINK> ). Therefore, we performed ancestral reconstructions using several models for each of the three types of reconstructions (nucleotide-based, amino acid–based, and codon-based). To determine which model of each type provided the best fit to the data, likelihood ratio tests (Navidi, Churchill, and von Haeseler 1991<$REFLINK> ; Yang, Goldman, and Friday 1994<$REFLINK> ) were carried out among pairs of nested models (table 1 ). Among nucleotide reconstructions, although GTR+Γ (Yang 1994<$REFLINK> ) gave the highest overall likelihood score, it was not significantly better than HKY+Γ (P = 0.33; Hasegawa, Kishino, and Yano 1985<$REFLINK> ), which in turn was significantly better than all simpler models tested. For codon-based reconstructions (Goldman and Yang 1994<$REFLINK> ; Muse and Gaut 1994<$REFLINK> ), the most parameter-rich model in which all the codon frequencies are allowed to vary, F61+Γ, was significantly better than simpler models tested. Among the amino acid models shown in table 1 , which differ only in the fixed substitution rate matrices, Jones+F+Γ (Jones, Taylor, and Thornton 1992<$REFLINK> ) gave the highest likelihood score. For all models tested, eliminating the Γ distribution, which accounts for among-site rate heterogeneity, resulted in a significantly worse fit to the data (HKY: 2ΔL = 2824.6, χ2() = 6.63, P < 0.001; F61: 2ΔL = 442.5, χ2() = 6.63, P < 0.001; Jones: 2ΔL = 448.8, χ2() = 6.63, P < 0.001). Not allowing amino acid frequencies to vary in the Jones+F+G model also proved to be significantly worse (2ΔL = 110.7, χ2() = 36.2, P < 0.001).
For the three best-fitting models, likelihood reconstructions of the ancestral archosaur rhodopsin were found to be in agreement at all but three amino acid sites (positions 213, 217, and 218), at which one of the three reconstructions differed from the other two. At position 213, the HKY+Γ model gave Ile instead of Thr; at 217, Jones+F+Γ gave Ala instead of Thr; and at 218, Jones+F+Γ gave Ile instead of Val. In designing the synthetic archosaur gene shown in figure 2 , the residue that agreed in two of the reconstructions was chosen; in fact, posterior probabilities for this residue tended to be higher than that for the alternative reconstructions (F61+Γ model: T213 (0.85), T217 (0.82), V218 (0.76); HKY+Γ model: I213 (0.60), T217 (0.86), V218 (0.75); Jones+F+Γ model: T213 (0.71), A217 (0.39), I218 (0.62)). Note that across all reconstructed sites, marginal posterior probabilities tended to be above 0.9 for the best-fitting models (fig. 2 , inset). On the basis of the crystal structure of bovine rhodopsin (Palczewski et al. 2000<$REFLINK> ), amino acid side chains at positions 213, 217, and 218 are expected to reside in the fifth transmembrane helix facing the bilayer lipid and might not be expected to significantly affect function, particularly in terms of spectral sensitivity and second-messenger activation. To test whether this is the case, we also synthesized a series of archosaur rhodopsin variants that contained the alternate amino acid reconstructions at the three sites (single mutants T213I, T217A, V218I, as well as a triple mutant incorporating all three substitutions).
The reconstructed ancestral archosaur rhodopsin amino acid sequence shown in figure 2 was used to design an artificial gene, which was synthesized in large fragments (168- to 230-mers) on a solid-phase oligonucleotide synthesizer. The synthesized fragments were amplified using the Pfu polymerase, cloned into the pCR-Blunt vector, pieced together using unique restriction sites, and cloned into a mammalian expression vector (pMT). This artificially synthesized ancestral archosaur gene was then transfected into monkey kidney (COS-1) cells, harvested, regenerated with 11-cis retinal in the COS cell membranes, solubilized, and purified (Ferretti et al. 1986<$REFLINK> ; Chang, Kazmi, and Sakmar 2002<$REFLINK> ).
The purified ancestral archosaur rhodopsin bound to 11-cis retinal to produce a stable pigment with a visible absorption maximum at 508 nm (fig. 3A ), which is redshifted from that of most mammalian and fish rhodopsins but within the higher end of the range of values reported for reptiles and particularly birds, which tend to have longer wavelength–absorbing rhodopsins (see fig. 1 ). Upon bleaching with light, the visible absorption peak shifted to 383 nm, which is characteristic of the active conformation of metarhodopsin II (inset, fig. 3A ). To determine if the light-activated conformation of the ancestral archosaur rhodopsin was functionally active, a fluorescence assay was used to measure guanine-nucleotide uptake by the heterotrimeric G-protein transducin. The photolyzed archosaur pigment activated transducin at a rate similar to that of bovine rhodopsin (86% normalized relative to bovine rhodopsin; fig. 3B ). Similar experiments were carried out on the ancestral archosaur rhodopsin variants (T213I, T217A, and V218I), which represent all possible alternate reconstructions. These variants showed similar results both in terms of spectral properties (λmax = 508 for all three) and transducin activation rates (83%, 74%, and 79%, respectively). A triple-replacement variant was also found to have spectral properties similar to the archosaur rhodopsin (λmax = 509).
These results indicate that the ancestral archosaur rhodopsin synthesized in the laboratory is able to activate the G-protein transducin in much the same way as bovine rhodopsin when assayed directly and that its spectrum is slightly redshifted. Moreover, at this level of divergence alternate amino acid reconstructions generated by different likelihood models, which were also synthesized and expressed, displayed similar functional characteristics. This indicates that archosaurs may have had a class of visual pigments that would support dim-light vision, which is consistent with the intriguing possibility that nocturnal, not diurnal, life histories may have been the ancestral state in amniotes (Gauthier 1994<$REFLINK> ), though further studies will be needed to clarify this issue.
Fossils preserved well enough to shed light on physiology and behavior are extremely rare (Ruben et al. 1999<$REFLINK> ; Fisher et al. 2000<$REFLINK> ). Attempts to amplify ancient DNA from exceptional samples preserved in amber or from dinosaur bone extracted from Cretaceous period coal beds have met with questionable success; in fact, material older than several hundred thousand years may not prove to be a reliable source of DNA, except under highly unusual circumstances (Hoss et al. 1996<$REFLINK> ). An entirely different approach is to use phylogenetic methods to infer ancestral sequences (Yang, Kumar, and Nei 1995<$REFLINK> ). One elegant study (Jermann et al. 1995<$REFLINK> ) recreated in the laboratory the molecular evolution of ribonuclease, specifically in the artiodactyl lineage, whose ancestor was estimated to have lived approximately 40 MYA. These types of approaches combine phylogenetic inference of ancestral gene structure with gene synthesis methods to obtain biological molecules that can be characterized in detail to provide a better understanding of the biology of ancient animals.
The vertebrate rhodopsin alignment used to infer the sequence of an ancestral archosaur rhodopsin has been deposited in the EMBL alignment database (Accession number ALIGN_000323). The synthetic archosaur rhodopsin sequence has been deposited in GenBank (Accession number AF310191).
Rodney Honeycutt, Reviewing Editor
Keywords: phylogenetic ancestral reconstruction rhodopsin biochemical evolution vertebrate vision G protein–coupled receptor
We thank Z. Yang, J. Bielawski, D. Campbell, R. L. Honeycutt, and three anonymous reviewers for their helpful comments and suggestions, and E. Marin for his help with the transducin fluorescence assay. This work was supported in part by the Howard Hughes Medical Institute, the Allene Reuss Memorial Trust, an NSF/Sloan Fellowship in Molecular Evolution (B.C.), and the Ellison Medical Foundation.