By sequencing genomes of taxa occupying key positions in the metazoan tree of life, it has become possible to infer when important animal gene families originated and proliferated 
. Sequence data alone, however, cannot yield insight into the functions and structures of ancient proteins or the processes by which their descendants evolved. Further, many gene families have diversified so extensively that comparisons of extant proteins from model organisms are insufficient to reveal which functions are ancestral and which are derived. In principle, it should be possible to reconstruct the history of a protein family by phylogenetically analyzing the underlying structural mechanisms that produce functional diversity among densely sampled members of the family. Such a strategy would be analogous to detailed studies of the evolution of animal development, which have revealed the deep homology of diverse morphologies in distant lineages and the mechanisms by which they evolved from common ancestral forms 
The members of the superfamily of nuclear receptor (NR) transcription factors, for example, are regulated in diverse ways—by ligands, postranslational modifications, and association with other proteins or DNA—depending on the cellular context 
. NRs have a modular domain structure, including a highly conserved DNA-binding domain (DBD) and a moderately conserved ligand-binding domain (LBD)—which in most receptors contains a ligand-regulated transcriptional activation function—along with extremely variable hinge and N-terminal domains. There is considerable diversity in the functions of NR LBDs, which can be roughly classified into three major groups. In one class, the LBD's transcriptional function can be activated by a specific hydrophobic ligand, such as a steroid, retinoid, or fatty acid; the ligand binds in a deep internal cavity, remodeling and stabilizing the LBD's conformation to generate a new binding surface for coactivator proteins, which increase transcription of nearby genes 
. The second class of NRs are ligand-independent transcription factors, often called “constitutive” receptors, the LBDs of which can adopt the active conformation and activate gene expression in the absence of a ligand or other modifications. Some members of this class lack the internal cavity and are not known to bind any ligands, whereas others do bind hydrophobic molecules, which up- or down-regulate their baseline activity 
. In the third class of NRs, the LBD lacks the capacity to interact with coactivators, so these receptors function primarily as transcriptional repressors that occupy NR response elements or dimerize with and thereby silence other NRs 
It is widely believed that the NR superfamily evolved from a ligand-independent transcriptional activator, with binding of different ligands gained independently in numerous NR lineages 
. The alternate view—that NRs evolved from a liganded ancestor, with ligand-dependence lost in the lineages leading to the ligand-independent receptors—has received little attention. These two hypotheses exemplify opposite views on the general nature of molecular evolution and the origin of complex functions. The hypothesis that the ancestral NR was ligand-independent implies that a complex molecular function—allosteric regulation of transcription by binding a ligand—evolved de novo many independent times, requiring evolution to repeatedly create novelty and complexity 
. In contrast, the hypothesis of a ligand-activated ancestor implies that evolution produced new functions primarily by subtle tinkering with a conserved ancestral mechanism 
, which allowed receptors to accommodate new molecular partners or lose dependence on those partners because of mutations that modified or degraded existing functions.
Several limitations have impeded rigorous inference about the ancestral NR's characteristics and the diversification of the superfamily. First, the root of the gene family phylogeny has been ambiguous, leaving unknown the location of the ancestor relative to its descendants. Second, the topology of the NR phylogeny has been uncertain, because of limited sequence sampling and/or use of outdated phylogenetic methods. Third, the functions of NRs in taxa branching near the root of the metazoan phylogeny have not been characterized. Finally, whether distantly related NRs with similar functions share homologous or convergent underlying mechanisms has not been determined. Recently acquired information—including genome sequences from basal metazoans and extensive data on NR structures and functions—along with improved algorithms for phylogenetic analysis of large datasets, now allow these barriers to be overcome. Here we report on biochemical, functional, structural, and phylogenetic analyses of the NR superfamily, which allow us to reconstruct the characteristics of the ancestral nuclear receptor and understand how the functional diversity of NR LBDs evolved.