The most striking feature of peafowl (Pavo) is the males' elaborate train, which exhibits ocelli (ornamental eyespots) that are under sexual selection. Two additional genera within the Phasianidae (Polyplectron and Argusianus) exhibit ocelli, but the appearance and location of these ornamental eyespots exhibit substantial variation among these genera, raising the question of whether ocelli are homologous. Within Polyplectron, ocelli are ancestral, suggesting ocelli may have evolved even earlier, prior to the divergence among genera. However, it remains unclear whether Pavo, Polyplectron and Argusianus form a monophyletic clade in which ocelli evolved once. We estimated the phylogeny of the ocellated species using sequences from 1966 ultraconserved elements (UCEs) and three mitochondrial regions. The three ocellated genera did form a strongly supported clade, but each ocellated genus was sister to at least one genus without ocelli. Indeed, Polyplectron and Galloperdix, a genus not previously suggested to be related to any ocellated taxon, were sister genera. The close relationship between taxa with and without ocelli suggests multiple gains or losses. Independent gains, possibly reflecting a pre-existing bias for eye-like structures among females and/or the existence of a simple mutational pathway for the origin of ocelli, appears to be the most likely explanation.
As no ornaments are more beautiful than the ocelli on the feather of various birds … they deserve to be especially noticed. [1, p. 132]
The wide diversity of colour and ornamental traits in birds has long been of interest to both scientists and non-scientists. Indeed, the ornamental traits of birds were instrumental to Darwin's development of the theory of sexual selection , where he devoted four chapters to the subject. The striking train of male peafowl, comprising elongated tail coverts decorated with ocelli (ornamental eyespots), had a special place within this discussion of avian sexual selection. Assuming gradual evolutionary change, Darwin recognized that there must have been species of birds with successive steps between ‘the magnificent ocelli of [peacocks] and the simpler ocelli or mere coloured spots of other birds’ (, p. 137).
To gather evidence of their gradual evolution, Darwin studied the striking ocelli present on male peafowls (Pavo sp.) in addition to similar structures found among species of two other genera in the same order (Galliformes) and family (Phasianidae): Polyplectron (peacock-pheasants) and Argusianus (argus pheasants). He immediately noted significant variation in the size, number and coloration of ocelli among these genera in addition to observing differences in the distribution of ocelli among taxa—Argusianus have ocelli located on the wing feathers and Polyplectron have ocelli distributed among different feather types (including tail, tail coverts, flight feathers, wing coverts and the mantle). Darwin also noted that females either lack ocelli (Pavo and Argusianus) or have smaller, less complex and/or less iridescent ocelli (Polyplectron) than males. It has since become clear that the ocelli of Pavo species are involved in sexual selection [2–4] and that male display of ocelli towards females during mating  also suggests a role for ocelli in sexual selection among Argusianus and Polyplectron. Indeed, males in all of these taxa prominently display ocelli towards females during mating displays , often with careful orientation to maximize impact of the ocelli . Darwin postulated that the ocelli of these taxa were homologous (using modern terminology), suggesting a single origin and then subsequent modification and elaboration via sexual selection.
Darwin's hypothesis predicts that modern, phylogenetic analyses should strongly support a close relationship among galliforms having ocelli. However, the inferred relationships among these taxa remain controversial. Although there is strong support for a peafowl clade (Pavo and Afropavo, the Congo peafowl) [7–10] and an argus clade (Argusianus and Rheinardia, the crested argus) , the relationships between these two clades and Polyplectron have varied (reviewed in ). A few studies unite the peafowl, argus and peacock-pheasants into an ‘ocellated clade’ [5,10], though with little support; a larger number of studies unite the peafowl and argus clades to the exclusion of the peacock-pheasants [9,11–13], whereas others do not unite any of these three clades . The position of Polyplectron is especially variable among studies , suggesting that large amounts of data may be necessary to resolve this question.
To further complicate matters, both the peafowl and argus clades contain one genus with ocelli (Pavo and Argusianus) and one genus lacking ocelli (Afropavo and Rheinardia), suggesting either a loss in Afropavo and Rheinardia or independent gains of ocelli in Pavo and Argusianus . Ocelli also occur in the earliest diverging members of the genus Polyplectron  but two derived species (Polyplectron inopinatum and Polyplectron chalcurum) exhibit reduced or no ocelli, suggesting loss of ocelli in this genus. Additionally, recent studies using a limited amount of mitochondrial data suggest Haematortyx (crimson-headed partridge, which lacks ocelli) might be closely related to Polyplectron [14,15]. If correct, this relationship provides another example of a highly elaborated, dimorphic group closely related to a species with little elaboration or dimorphism, further emphasizing the lability of secondary sexual traits in galliforms [14,16]. The difficulties in resolving phylogenetic relationships among the ocellated galliforms combined with the potential loss of ocelli within genera have made it unclear whether ocelli are homologous (with subsequent loss in some taxa) or whether ocelli arose independently in each ocellated genus.
Here, we ask whether the ocellated galliform taxa form a clade and determine the closest relatives of the enigmatic genus Polyplectron. To accomplish this, we used a supermatrix analysis including all galliform species with available sequence data (233 species) to select 15 target species that represent: (i) taxa with ocelli, (ii) the potential sister taxa of any ocellated genera and (iii) several appropriate outgroup species (electronic supplementary material, table S1). We inferred evolutionary relationships among these 15 target species using high-throughput sequencing of ultraconserved elements (UCEs) and their flanking DNA. Finally, we reconstructed the ancestral condition of ocelli to examine whether ocelli are likely to have evolved once or multiple times.
2. Material and methods
(a) Supermatrix analyses to identify target taxa
To identify all possible sister taxa of the ocellated taxa, we consulted Wang et al. , who reviewed previous galliform phylogenies, and we used available data to construct a mitochondrial supermatrix of CYB, ND2 and 12S rRNA sequences from GenBank for Galliformes (electronic supplementary material, table S2). We supplemented those data with mitochondrial sequences collected by different methods, including high-throughput sequencing  and PCR amplification from toe pads (see  for methods). The final matrix included 233 species, corresponding to 83% of Galliformes and 91% of Phasianidae (the family that includes the ocellated taxa).
Using this mitochondrial supermatrix, we performed unpartitioned and partitioned (by gene and codon position) maximum-likelihood (ML) analyses using RAxML v. 7.3.0 . We analysed the mitochondrial DNA using the best model (GTR + I + Γ), identified using the Akaike information criterion (AIC) in Modeltest v. 3.06  and assessed support using 1000 bootstrap replicates. We also used Bayesian inference (BI) to assess support, conducting analyses in MrBayes v. 3.1.2  on the CIPRES Science Gateway  using the GTR + I + Γ model. We ran the analyses for 107 generations using the standard settings (two runs, each with four heated chains) and we sampled every 1000 generations. The first 25% of the sampled trees were discarded as burn-in.
(b) Ultraconserved element data collection
We selected 15 species for UCE enrichment using a combination of literature searches and the trees inferred above. These taxa included representatives of the three ocellated genera, putative sister taxa and multiple outgroups. We collected UCE data using a modification of the Faircloth et al.  approach. Briefly, we prepared Nextera sequencing libraries using the manufacturer's protocol (Illumina Inc., San Diego, CA, USA). We pooled libraries into groups of eight taxa, enriched each library pool for 5060 UCE loci (targeted by 5472 probes), amplified the enriched pools using limited-cycle PCR (18 cycles), quantified the resulting pools using qPCR (Kapa Biosystems) and sequenced the enriched libraries using a single Illumina HiSeq 2000 lane (PE75 reads generated at the UC Irvine Genomics High-Throughput Facility). We conducted de novo assembly using Velvet , matched the contigs to defined UCE loci using PHYLUCE v. 1.1 , and added UCE sequences from the Gallus gallus and Meleagris gallopavo genome assemblies [25,26]. We excluded loci having missing data for any taxon and aligned the sequences using MAFFT . We examined alignments by eye in Geneious Pro (v. 5.6.6) and removed likely assembly or alignment errors at the 5′ or 3′ ends prior to subsequent analyses.
(c) Phylogenomic analyses of ultraconserved element data
The 15-taxon dataset comprised 1966 UCE loci after excluding loci with missing data in any taxon, producing a completely sampled matrix. The average locus length was 409 bp (range = 344–746 bp) and the concatenated alignment contained 803 901 sites (17 559 were parsimony informative). We conducted unpartitioned and partitioned ML analyses and BI using the GTR + I + Γ model (the best-fitting based on the AIC ). Importance values of the base composition parameters for individual UCEs were much higher than those for other parameters (electronic supplementary material, figure S1) and GC content was the axis for which parameter estimates exhibited the greatest variation (electronic supplementary material, figure S2). Therefore, we partitioned the UCE data based upon GC content, using five bins (30–38%, 38–46%, 46–54%, 54–62% and 62–70% GC) containing varying numbers of loci.
We implemented the SMRT-ML  method to estimate the species tree using the UCE data. SMRT-ML is a method for analysing concatenated data that is consistent when gene tree topologies differ due to incomplete lineage sorting under the multispecies coalescent. This allowed us to use of all of the data and avoid including poorly resolved gene trees as is done in a standard coalescent-based analysis. We chose this approach because the coalescent-based analysis in a previous UCE study  had limited resolution, possibly reflecting, at least in part, the inclusion of poorly resolved gene trees. To implement SMRT-ML, we wrote a perl script that identified ML trees (using GTR + I + Γ without partitioning in RAxML v. 7.2.8 ) for each possible rooted triple and then combined quartettes using the modified MinCut method . We estimated support using 100 bootstrap replicates generated prior to running the SMRT-ML script.
To provide a taxon-rich dataset with the power to reconstruct the phylogeny of the ocellated taxa, we combined the 233-taxon mitochondrial matrix with the 15-taxon UCE dataset. This dataset included all eight of Polyplectron and Pavo, two (of three) species of Galloperdix, as well as Argusianus, Rheinardia and Haematortyx. The optimal tree was estimated using RAxML and the GTR + I + Γ model with 10 randomized starting trees. Species represented only by mitochondrial data had longer branch lengths than those represented by both mitochondrial and UCE data. Therefore, the branch lengths for the combined UCE + mitochondrial topology were re-estimated using only mitochondrial data.
Presence and absence of ocelli was scored as a binary trait. Although the ocelli in P. inopinatum are reduced (simpler than in some of the other species), they were scored as present. To reconstruct ancestral states using ML, we used Mesquite 2.72  and the StochChar package  to determine whether a two-rate model (gains ≠ losses; AsymmMk) was significantly better than a one-rate model (gains = losses) with a likelihood ratio test. After identifying the best model (the AsymmMk model), we reconstructed the evolution of ocelli using the StochChar package.
(a) Identification of genera related to ocellated taxa
The ML and BI phylogenies from the 233-taxon mitochondrial dataset strongly supported both a peafowl (Pavo–Afropavo) clade and an argus (Argusianus–Rheinardia) clade (figure 1 and the electronic supplementary material, S3), as expected. We also found strong support for an expanded peacock-pheasant clade comprising Polyplectron, Galloperdix and Haematortyx. Similar to previous studies [5,7,16], the mitochondrial data did not resolve relationships among the ocellated clades with high support (figure 1 and the electronic supplementary material, S3). Based on these results, we sampled UCE loci from Haematortyx and Galloperdix along with the ocellated taxa. To these taxa, we added additional phasianid taxa (n = 7) that were sufficient to test whether the topology for the ocellated taxa corresponded to that in any recent study (reviewed in ) as well as a non-phasianid outgroup.
(b) Resolving relationships among genera using ultraconserved elements
In contrast to previous studies and mitochondrial data (figure 1), which have been unable to resolve relationships among the ocellated taxa, we found a strongly supported clade uniting the ocellated taxa in both the ML and BI analyses of the UCE dataset (figure 2). As we found with the mitochondrial data, analyses of the UCEs suggested that Polyplectron, Haematortyx and Galloperdix form an expanded peacock-pheasant clade, united with the peafowl and argus clades. Many previous studies, including other studies with limited numbers of nuclear loci, have failed to support monophyly of the clade that contains Pavo, Argusianus and Polyplectron . This raised the question of whether our support for this clade reflects the change in taxon sampling, which breaks up the long branch to Polyplectron. However, the optimal tree united the three clades even when we excluded Haematortyx and Galloperdix (electronic supplementary material, figure S4), suggesting the topological differences among previous studies reflect the use of insufficient amounts of data. Finally, we note that the SMRT-ML tree also provided strong support for the clade comprising the ocellated taxa (figure 2), strongly suggesting that this clade is present in the species tree.
Multiple hypotheses can explain the distribution of ocelli (figure 3). However, ancestral state reconstruction using likelihood was equivocal regarding the presence or absence of ocelli in the common ancestor of Pavo, Argusianus and Polyplectron (electronic supplementary material, figure S5). In spite of this, our phylogenetic analyses do provide a strongly supported hypothesis of relationships among these taxa that can be used to examine alternative models of evolution of ocelli (as discussed below).
(a) Relationships among ocellated taxa
Our results provide strong support for a clade that includes all three genera of galliforms (Pavo, Argusianus and Polyplectron) that exhibit ocelli. However, we found that the clade also includes additional genera that lack ocelli. In addition to the expected sister relationships between Pavo–Afropavo and Argusianus–Rheinardia, we also found that Polyplectron is closely related to two non-ocellated genera: Galloperdix and Haematortyx. Two recent studies have suggested Haematortyx might be related to Polyplectron based upon limited mitochondrial data [14,15], but the close relationship of Galloperdix with these two genera was unexpected based upon the literature. Although Galloperdix and Haematortyx do exhibit sexual dimorphism, both genera have typically been considered partridges. By contrast, the other genera in this clade, including those without ocelli, are pheasant-like taxa that have specialized feathers and ornamentation.
(b) Evolution of ocelli
The simplest modern interpretation of Darwin's  hypothesis, that the ocelli of Pavo, Argusianus and Polyplectron are homologous, would suggest the existence of a clade comprising the ocellated taxa and excluding other galliforms. This simple interpretation was falsified by the support for close relationships between ocellated (Pavo and Argusianus) and non-ocellated (Afropavo and Rheinartia) taxa [5,7]. We further falsified this simple hypothesis when we found strong support for placing Galloperdix and Haematortyx as successive sister groups of Polyplectron. However, Darwin's  hypothesis does not preclude the inclusion of other taxa in this clade. Previous evidence for the loss of ocelli in P. chalcurum  raises the possibility that ocelli are homologous (i.e. there was a single gain) but that the single gain may have been followed by multiple losses.
With the potential for both loss and gain of ocelli, there are three different hypotheses that are consistent with our phylogeny (figure 3). Hypothesis 1 assumes a single gain, consistent with Darwin's  hypothesis, but then requires five subsequent losses (six total evolutionary steps). Hypothesis 2 is an alternative that requires two independent gains and at least three losses, resulting in five steps. However, the most parsimonious model of ocellus evolution (hypothesis 3) requires only four steps, three independent gains in each of the three ocellated genera and a single loss within the genus Polyplectron. Darwin's  hypothesis (our hypothesis 1) might be favoured if losses occur much more frequently than gains. The relatively short terminal branch length for P. chalcurum  indicates that loss of ocelli can occur relatively rapidly. By contrast, the rate of gain is less clear. If hypothesis 1 is correct, ocelli would have to be gained along the very short branch uniting the ocellated clade (figure 2). By contrast, the hypothesis of multiple origins would allow gain along the relatively long branches leading to each ocellated genus (figure 2), providing few constraints upon the rate of gain for these characters.
Given that a simple likelihood reconstruction does not completely address the evolution of ocelli, it is clearly desirable to incorporate other types of biological information to understand the evolution of ocelli. There are differences in the appearance and distribution of ocelli among feather types in Pavo, Argusianus and Polyplectron. Indeed, Darwin  stated that the existence of single ocelli in Pavo but paired ocelli in Polyplectron (figure 3; though we note that paired ocelli only occur on the tail feathers) made him question the hypothesis that the traits were related. Although homologous traits typically vary in appearance among species, the location of the ocelli also varies among the taxa of interest in this study (tail coverts in Pavo, greatly elongated secondaries in Argusianus, and broadly distributed on the body in Polyplectron; figure 3). These differences in distribution could be suggestive of multiple origins. Indeed, given the differences among taxa in the distribution of ocelli, the hypothesis of a single origin would require either (i) an ancestor with ocelli on multiple feather types (similar to most Polyplecton species) combined with loss on most feather types in Pavo and Argusianus or (ii) an ancestor with ocelli on a single feather type combined with shifts among feather types (i.e. the development of ocelli on a novel feather type combined with the loss of expression on the original feather type). Thus, even if Darwin's  hypothesis of homology were correct (hypothesis 1), the evolution of galliform ocelli must be more complex (and therefore even less parsimonious) than a single origin and five losses.
There are multiple explanations for the loss of ocelli. Kimball et al.  presented strong evidence for the loss of ocelli within the genus Polyplectron, noting that both the species that has lost ocelli and a second species with simplified ocelli have montane distributions, unlike the lowland distribution of the ocellated Polyplectron species. Reduced male ornamentation with increased altitude has been observed in other birds , possibly reflecting a reduction in the strength of sexual selection and/or the increased cost of trait production due to resource limitation. However, the other members of the larger clade that lack ocelli (Afropavo, Rheinardia, Galloperdix and Haematortyx) inhabit lowlands. Therefore, even if a single origin of ocelli is correct, one cannot invoke altitudinal shifts to explain loss of ocelli in these taxa. Nonetheless, there may be additional, as yet unknown, factors that could have an impact upon the strength of natural and/or sexual selection leading to the loss of ocelli in these species.
The most parsimonious hypothesis, which invokes multiple independent gains of ocelli (and a single loss within Polyplectron, see figure 3, hypothesis 3), is potentially of greater interest. Modified feathers and featherless regions on the head of many phasianid species highlight the eyes during displays. In addition, female preference for traits that emphasize the eye have been identified (e.g. [32–34]), suggesting there may be a general female preference for eye-like structures . Thus, female phasianids may have a pre-existing sensory bias  for eye-like structures. Any genetic mutations in males that lead to eye-like structures or exaggerate the prominence of those structures might be favoured by sexual selection. If there are relatively few mutational events necessary to evolve ocelli (potentially expanding simple spots to a complex ocelli), then it is possible that this pathway has occurred repeatedly on different feather types among these taxa.
Phylogenomic methods, such as sequence capture , make it possible to obtain large amounts of data to infer phylogenetic relationships with a high degree of confidence. Although previous galliform phylogenies, including multi-locus studies, have been equivocal regarding the relationships among ocellated taxa, the UCE data provide strong support for a clade that includes the three ocellated genera. The data also strongly support the inclusion of taxa without ocelli within the clade. Despite the well-resolved phylogeny, ancestral state reconstruction remained equivocal. Given valid explanations for both multiple losses as well as multiple gains, Darwin's  hypothesis cannot be fully addressed with phylogeny alone. However, considering other types of information suggests that multiple gains may be likely. Improvements in our understanding of feather development  may allow an examination of the developmental pathways underlying the presence of ocelli (versus spots and other feather features) to further our understanding of the evolution of ocelli. We believe that this is likely to be a fruitful next step now that the phylogeny of this group has been established with confidence.
DNA sequences generated as part of this study have been submitted to GenBank (accession numbers: KJ968313–KJ993870), while the raw sequence reads have been submitted to the Sequence Read Archive (BioProject: PRJNA252000; accession: SRP043045). Sequence alignments, trait files, tree files and the SMRT-ML script have been submitted to DRYAD (doi:10.5061/dryad.66942).
This work was supported by the US National Science Foundation (grant no. DEB 1118823 to R.T.K. and E.L.B.).
We thank the Kimball-Braun laboratory and two anonymous reviewers for helpful comments on earlier versions of this manuscript. We thank the Louisiana State University Museum of Natural Science for generous donation of tissues. Hannah Owens kindly supplied photographs of Argusianus and Polyplectron.
- Received April 4, 2014.
- Accepted June 24, 2014.
- © 2014 The Author(s) Published by the Royal Society. All rights reserved.