The complete 1,143 nucleotides of the cytochrome b gene were retrieved in seven overlapping fragments (see primers in Table 1), including the 338 base pairs section between positions 14,312 and 14,649 (using Ovis as reference sequence) already retrieved in a previous work 11 and a section of 85 bp replicated in Oxford (between positions 14,399 and 14,483) (Table 2). Two discrepancies with the previous consensus sequences were observed, a T in position 14,635 and a T in 14,638. The first one was already reported as heteroplasmic in the cloned PCR product in 11, where it was present in five up to ten clones sequenced. The latter was not found in our previous study; therefore, it can correspond to a silent polymorphism within the Myotragus population or to a DNA damage involving a rare A to T change in that particular PCR reaction. All cytochrome b fragments were routinely cloned (not shown), although very few heteroplasmies were detected in the direct sequencing, attributable to both the exceptional preservation of the sample and the short lengths of the fragments; consequently, the error rate in the clones (number of nucleotide differences/1,000 base pairs) was very low (< 2 per thousand base pairs). Considering the overlapping of the primers and the fragment replicated, 404 bp (around 35% of the cytochrome b gene) have been determined from more than one PCR. It cannot be discarded that DNA damage could have affected some positions, although it is unlikely that this would significantly alter the phylogenetic inferences.

The putative presence of nuclear mtDNA insertions is very unlikely, since we proceed designing the L primers for the next fragment in the sequence already retrieved from the previous fragment, and the H primers in a consensus Caprinae sequence. The 12S gene sequence (305 base pairs) was retrieved in three overlapping fragments (see primers in Table 1). The sequences found in 12S showed several differences from those of extant Caprinae, a fact that points to its authenticity.

The PCR of the 28S nuclear gene yielded a very faint band around 140 base pairs and was subsequently cloned and sequenced. In addition, DNA from tissue samples from domestic goat (Capra aegagrus), southern chamois (Rupicapra pyrenaica) and domestic sheep (Ovis aries) was extracted in a separated laboratory, and the same 28S fragment was determined. The 96 base pairs sequence obtained from Myotragus is similar to that found in other Caprinae (Table 3), and clearly different to the human one and to the cow, a suggested source of contamination because of the use of BSA (bovine serum albumine) in ancient DNA amplifications. Moreover, no other Bovids, extinct or living, had been analyzed in the same laboratory when the extraction, amplification, cloning and sequencing of the Myotragus genes was undertaken; therefore, the 28S sequence seems to be endogenous of Myotragus. The cloning of the 28S PCR product (Table 3) showed that two of the sequences (about 20% of the clones) were human contaminants; this accounts for the noise observed in the direct sequencing of the PCR product. Most likely, the human DNA comes from handling of the Myotragus bones during its excavation and posterior morphological study. Nevertheless, the error rate in the clones is relatively low (3.07 per thousand base pairs), a figure within the range found in some modern specimens and well preserved ancient remains, like the moas 20.

Variation in length is characteristic of 28S gene; size variations are due to expansions or contractions of variable segments in 10–12 positions within the gene, where variation does not interfere with ribosomal function 21. The sequences were aligned by eye; due to the high interspecific variation in 28S, no definite alignment is possible and therefore, the phylogenetic information obtained from this gene is limited. However, the retrieval of a nuclear gene indicates the quality of the DNA of the sample used. Furthermore, it opens new possibilities of research for assessing the phylogenetic relationships of Myotragus and for retrieving nuclear genes with phenotypic implications.

We included Pantholops hodgsoni in the phylogenetic analysis since previous analysis as well as ours indicate that this species is clearly associated to other members of the subfamily Caprinae. As an outgroup, we used 14 members of the tribes Alcephalini and Hippotragini (see Methods) previously detected as the closest members within the Bovidae 1722. The cytochrome b and cytochrome b + 12S Bayesian and maximum-likelihood trees (Figure 1) showed a topology roughly similar to that found in previous studies on Caprinae phylogeny (e.g., 22). Recently, it has been reported that the Budorcas cytochrome b sequence available at GenBank was in fact a chimaera that included a fragment of Ovis sequence, and a new sequence for this species was provided 17. Crucially, in previous works Myotragus formed a quite stable clade with Budorcas and Ovis that is not maintained anymore, with Budorcas now clustering in a different position.

Examination of the cytochrome b trees (Figure 1a and 1b) shows that some of the clades previously found are present in these trees, specially Myotragus+Ovis (67.5% in the maximum-likelihood trees), Capricornis+Ovibos+Naemorhedus (100% in the maximum-likelihood tree) and Hemitragus+Pseudois+Capra (94.6% in the maximum-likelihood tree). The position of Hemitragus, that precludes the monophyly of the genus Capra, is problematic, as some authors have already noticed 18. Some species show unstable positions in the trees, particularly the genera Budorcas, Ammotragus, Pantholops, Rupicapra and Oreamnos. The problematic position of these taxa has been already reported on extant species studies (see, for instance 2223). However, our tree, although still with low bootstrap values, has been able to better resolve the phylogenetic position of Myotragus, which is basal to the Ovis clade; the support value for the Myotragus+Ovis grouping is now 0.46 in the Bayesian tree and the bootstrap value 67.5% in the maximum-likelihood tree, respectively 11.

The trees with the cytochrome b + 12S fragment (305 base pairs) did not significantly improve the phylogeny (Figure 1c and 1d); the Capricornis+Ovibos and Hemitragus+Pseudois+Capra clades are well supported by bootstrap analysis, while the support of the cluster of Myotragus+Ovis is reduced with respect to the cytochrome b sequence alone. This discrepancy can be attributed to the small number of informative positions added by the 12S fragment together with the reduced number of species that could be used due to the unavailability of the 12S sequence in many species.

The complete resolution of the Caprinae phylogeny cannot be achieved with the cytochrome b gene alone, as some species of the subfamily (specially of the genera Budorcas, Oreamnos, Rupicapra, Ammotragus and Pantholops) still have an unstable position. It is likely that the lack of resolution in the basal branches is due to the existence of a very quick initial radiation in the Caprinae subfamily. To test this, we performed a likelihood ratio test for zero branch-length of all branches of the tree, that showed that there are nine branches with p > 0.05 (and therefore, not significantly different from 0). Five of them belong to the basal branches of the Caprinae tree, thus supporting the hypothesis of a very fast initial radiation of this group. Additionally, the plot of grow of lineages of an ultrametric tree (see below) also supports this fast initial radiation.

Being Myotragus a small caprine, its long branch in the maximum-likelihood tree previously detected with a cytochrome b fragment 11 was attributed to an earlier age of first reproduction and a shorter generation time in Myotragus than in other wild caprines 11. This long branch is also present in the trees of the complete cytochrome b (Figure 1a and 1b). However, recent estimates indicate that Myotragus was actually bigger than previously believed, with a weight ranging between 15 and 25 kg for the smallest adult specimens and between 40 and 60 kg for the larger specimens 24. Therefore the reason for this long branch cannot be due to its presumed extreme reduced size (and the consequences that this involves). On the other hand, the hypothesis that tries to explain differences in evolutionary rates as due to differences in metabolic rate or generation time (both correlated to body mass) was not supported in the analysis of cytochrome b evolutionary rates of a wide set of mammalian species 25. It is thus more likely that this relatively long branch of Myotragus in the Caprinae tree is mainly a stochastic phenomenon.

Due to the endemicity of Myotragus 2, it can be hypothesized that this lineage diverged from the continental species 5.35 million years ago (Mya), when the desiccation of the Mediterranean ended 1314 and the Myotragus ancestors became isolated in the Balearic Islands. In principle, this would be a minimum age for the node in the tree, but we consider likely that the isolation of the Balearic Islands was a vicariant event responsible for the split of the Myotragus and continental lineages, so that we can use this date as a fixed age in the Caprinae cytochrome b tree (Figure 2). The dating results for the maximum-likelihood and Bayesian trees are very similar and only the former are shown. We have calculated ultrametric trees using the Langley-Fitch method 26 as well as the penalyzed likelihood method with several smoothing factors 27 implemented in the r8s program. The cross-validation procedure to explore the fidelity with which these different methods explain the branch length variation in the tree indicates that the Langley-Fitch method is slightly better for our data set and therefore we used this method in the calculations described next (although dates obtained were very similar using penalyzed likelihood with a log₁₀ of the smoothing factor of 2; not shown). In addition, we have performed a parametric bootstrap procedure, in which we simulated alignments from the maximum-likelihood tree and re-calculated dates from the same maximum-likelihood topology with branch lengths optimized for each simulated alignment. This allowed us to evaluate the stochastic errors of date estimates associated to sampling a finite number of alignment positions 28. In a different parametric bootstrap procedure, we reconstructed new trees from each simulated alignment and used them for time estimation (instead of only the original topology) and therefore errors associated with finite number of positions as well as with tree reconstruction imperfections are estimated 28. The standard deviations and 95% confidence intervals for most date estimates were very similar for both parametric bootstrap procedures (Table 4), indicating that most of the errors in the present data set are associated to the finite sample sequence rather than to tree errors. In the following, standard deviations given refer to the parametric bootstrap using only the original topology for calculating dates.

The ultrametric trees calculated by the Langley-Fitch method applied to the cytochrome b maximum-likelihood tree showed that the separation of the most divergent lineages within Caprinae (Pantholops and the rest of species) occurred 6.2 ± 0.4 Mya. These dates are more recent than a proposed Caprinae radiation 11 Mya based on different molecular calibrations (see Introduction) 1718. There is no general consensus from fossil data on the date of the start of the Caprinae radiation, that ranges from around 14 Mya 12 to a more recent Late Miocene origin 29, which would be more consistent with our results.

After the first split, the separation of the most basal lineages took place around or before 5.35 Mya (Figure 2), and therefore we can estimate that the main radiation of caprines occurred in less than one million years of evolution. A lineages through time plot, that reflects the number of lineages in each split point of the tree (bottom of Figure 2), indicates a fast growing of lineages during the first million years of evolution in comparison with the flatter slope in the rest of the tree. Although a few species are missing in this tree, all genera are covered and it is likely that the tree with all species will reveal the same clear steep slope at its base. This truly supports the existence of a quite fast initial radiation that led to the differentiation of nine clades (including the Myotragus lineage) that are at the base of all extant genera. In addition, this explains the difficulties we face in reconstructing the phylogenetic tree of this group.

Interestingly, the separation of the two chamois species (Rupicapra) and the beginning of the radiation of the main clade of goats (separation of the Capra ibex and Capra pyrenaica clade from the rest of species) occurred at 1.6 ± 0.3 Mya and 1.5 ± 0.2, respectively, according to the calibrated cytochrome b tree. This would imply that the beginning of the Pleistocene glaciations promoted the diversification of the main lineages of Caprinae species living in the Alps and other Eurasian mountainous regions. The deep split observed between the alpine and southern chamois, at 1.6 ± 0.3 Mya, is quite remarkable, since these two species have been considered subspecies until recently 30, and other molecular studies have given much more recent dates for the separation of these two species using restriction fragment length polymorphism of mitochondrial DNA 31 or microsatellites 32. Their split is even older than that between Capra pyrenaica and Capra ibex, a pair of species with a similar geographic disjunction than the two chamois and traditionally recognized as different species, that separated 0.6 ± 0.1 Mya according to our data. To test whether the deep split observed between the two chamois was caused by particularly divergent cytochrome b haplotypes of chamois used in this study, we performed the calibrations with all sequences of chamois in databases, including shorter ones (five for R. pyrenaica and six for R. rupicapra), obtaining very similar dates (not shown). To analyze if the date of this split depended on the particularly long branch of Myotragus, we manually reduced this branch to the half, but we obtained again similar dates for all splits in the trees (1.7 Mya for the split between chamois). This is due to the smoothing of rates performed by the calibration methods used, that correct for the higher evolutionary rate of the Myotragus lineage. To account for the possibility that Myotragus could be more basal in the tree, we manually positioned Myotragus just after the separation of Pantholops. Then the split date of the two chamois was reduced, but only up to 1.4 Mya. This is in agreement with the parametric bootstrap analyses (Table 4), that indicate that the tree topology is not crucial in this calculation. This is surely due to the small internal branches involved in all different topologies arranging the basal species of the tree. Other works where several topologies were used for calculating dates have also shown that age estimates were largely insensitive to different phylogenies 28.

Finally, if Myotragus and Ovis had diverged in fact more recently and the ancestors of Myotragus had reached the Balearic Islands via a transmarine colonization after these islands became disconnected from the continent, as proposed for other vertebrates 32, all the dates in the tree could be more recent; however, this hypothesis is extremely unlikely in Myotragus from the paleontological evidence 24. On the other hand, the use of evolutionary rates calculated from other sources, for example, the separation of humans and chimpanzees 5 – 7 Mya 3435 combined with the genetic distance between their cytochrome b sequence, would lead to a separation of Myotragus and Ovis clades at 7.2 Mya. Although the use of an external calibration could be problematic, this reference would not support a transmarine colonization of Myotragus ancestors. Thus, the calibrated cytochrome b trees indicate that alpine and southern chamois have been separated much longer than previously thought, which would help explain the anti-hybridization mechanisms currently operating 30.