Turtles (order Testudines) represent one of the most anatomically peculiar vertebrate groups. Their highly derived morphology relative to other vertebrates arose through profound structural changes associated with the origin of the shell [1
]. Turtles have been described as having a conspicuously modified reptile body plan, and termed 'hopeful monsters', representing a successful phenotypic mutant with the potential to establish a new evolutionary lineage [2
]. These morphological adaptations make it difficult to compare turtles with other organisms and to establish the polarity of numerous anatomical characters as either being ancestral or derived. Because turtles represent one of the major groups of amniotes, resolving their phylogenetic position would fill an important gap in the evolutionary history of vertebrates, and contribute to the understanding of how such a key innovation as the turtle shell originated and which underlying genes are involved in its development [8
The phylogenetic relationships of turtles within the amniotes have puzzled scientists for more than a century. Turtles have been classified both as basal to all other reptiles (including birds) and as nested within them, implying two radically different perspectives from which to interpret the evolution of morphological, developmental, molecular, or ecological data. The classic view [9
] places turtles as the sister group to all other reptiles, mostly on the basis of the lack of temporal fenestration in the skull, a character considered as being ancestral for Reptilia. This view reflects the traditional dichotomy of reptiles as Anapsida (lacking temporal fenestration) or Diapsida (with two temporal fenestrations). However, cladistic studies of morphological datasets have generated conflicting results, supporting both an anapsid [11
] and a diapsid [14
] affinity of turtles, depending on the fossil sampling considered and the morphological matrices used. Harris et al.
] showed how the morphological characters used to assess the phylogenetic placement of the turtles within the tree of amniotes can lead to conflicting signals, and suggested, given these difficulties, that the answer to this long-standing controversy would most probably come from molecular data.
However, the use of molecular data has not yet settled the debate, as it has also provided somewhat conflicting results. In the large number of publications on the topic over the past decade, turtles have been grouped with Archosauria (birds and crocodiles) in most studies [18
], but have also been grouped with crocodiles [23
] or sometimes with Lepidosauria (tuatara, lizards, and snakes) [27
]. The causes of the conflicting signals and/or lack of resolution obtained in most studies have been attributed to the limited number of genes considered, poor taxon sampling, substitution rate heterogeneity among genes and among taxa, and saturation or selection occurring at some of the markers [28
]. Moreover, statistical tests performed to evaluate alternative topologies based on these early molecular sequence datasets usually failed to reach significance, probably because of the reduced number of genes included, but also possibly because of heterogeneity in gene trees.
With the advent of genomic data, the comparative datasets increased in size, but the issue of turtle phylogeny remained unresolved. The first investigation of genome structure and composition in reptiles identified a similarity in genomic signatures between turtles and crocodiles [29
]. A recent multigene study offered the first convincing support for the grouping of turtles and archosaurs [30
], but this result was contradicted by a newer study based on the distribution of microRNAs (miRNA), which strongly suggested an alternative turtles plus lizards clade [31
]. Finally, a recent phylogenomic study based on reptile transcriptomic data did not find compelling support to distinguish between turtles plus crocodiles and turtles plus archosaurs, despite including a large number of genes [32
]. In that study, the analysis of the largest dataset strongly supported a topology with the turtle as the sister group to the crocodile, whereas analyses after removing potential paralogs favoured a turtle plus archosaurs clade, albeit with reduced statistical support [32
In the present study, we used a phylogenomic approach [33
] to resolve the position of the turtles within the amniotes, and estimated the time of their origin using a dataset comprising 248 nuclear protein-coding genes for 16 vertebrates. We applied phylogenetic reconstruction methods and models of sequence evolution, explicitly accounting for substitution rate heterogeneities among taxa and among genes, and maximum likelihood (ML) species tree analyses accounting for gene tree discordance. We also used various relaxed molecular clock Bayesian approaches to reconstruct a timescale for the evolutionary history of the amniotes.