Institutional Abbreviations

BMS, Buffalo Museum of Science, Buffalo, New York; FMNH, The Field Museum of Natural History, Chicago, Illinois; GMNH, Gunma Museum of Natural History, Tomioka, Gunma Prefecture, Japan; KMNH, Kitakyushu Museum and Institute of Natural History, Kitakyushu, Japan; MACN, Museo Argentino de Ciencias Naturales, Buenos Aires, Argentina; NSMT, National Science Museum, Tokyo, Japan; UCMP, University of California Museum of Paleontology; USNM, National Museum of Natural History, Smithsonian Institution, Washington, D.C.; UWBM, Burke Museum of Natural History and Culture, Seattle, Washington; UWGM, University of Wyoming Geological Museum, Laramie, Wyoming; WSGS, Wyoming State Geological Survey, Laramie, Wyoming.

Taxon Sampling

Though recent molecular and morphological analyses [4], [26], [27], [29] of higher-level avian phylogeny generally agree on the taxonomic content of the waterbird clade (with several noteable exceptions such as Phoenicopterus, Podiceps, and Phaethon), the fact that: 1) relationships within the waterbird tree are so contentious, 2) Pelecaniformes may not be monophyletic, and 3) fossil pelecaniforms may not actually be closely related to their purported extant pelecaniform families; necessitated a broad taxonomic sampling scheme that included diverse members from throughout the waterbird clade. Accordingly, 57 waterbird taxa were included in the analysis, with the following families represented: Gaviidae, Podicipedidae, Spheniscidae, Procellariidae, Diomedeidae, Hydrobatidae, Pelecanoididae, Phoenicopteridae, Ciconiidae, Ardeidae, Threskiornithidae, Balaenicipitidae, Scopidae, Phaethontidae, Pelecanidae, Fregatidae, Sulidae, Anhingidae, and Phalacrocoracidae (Appendix S1). Eudromia elegans (Tinamidae) and Gallus gallus (Phasianidae) were utilized as outgroups to root phylogenetic trees.

An important caveat is that the taxonomic scope of the present analysis is limited to the waterbird clade. Thus, the current dataset is not designed to test the global relationships of any included taxa within Aves. This is not a major problem for most included taxa, whose status as a member of the waterbird clade is uncontroversial. However, for taxa such as Phaethon, Podiceps, and Phoenicopterus, which have been recovered in previous phylogenetic studies outside of the waterbird clade, often as closely related members of a clade variably termed ‘Metaves’ [19], [27], [29], the issue of taxonomic scope is more of a concern. Accordingly, it is most appropriate to view the current analysis as a rigorous test of the relationships of these taxa if they are indeed waterbirds, and as uninformative of their relationships if they belong outside of the waterbird clade.

Limnofregata azygosternon was first described by Olson [35] as a member of the pelecaniform family Fregatidae, and a possible direct ancestor of modern Fregata. The holotype (USNM 22753), and all subsequently referred specimens with the exception of one (USNM 447002), were collected from the early Eocene Fossil Butte Member of the Green River Formation, which radiometric dating indicates is 51.97+/−0.16 Myr [58]. In 2005, a second species, Limnofregata hasegawai, was described, along with additional new material of L. azygosternon [59]. The new species is virtually identical to L. azygosternon, and differs from it only in its overall larger size, and proportionally longer rostrum [59]. As the two species of Limnofregata currently recognized [59] differ only in relative size and proportion of the rostrum, and not in any discrete anatomical characters, observations based on specimens from both L. azygosternon and L. hasegawai were lumped, and Limnofregata was coded as a single OTU in this analysis. Morphological characters were coded for Limnofregata based on first-hand examination of the majority of specimens, and were supplemented by published descriptions [35], [59] where necessary. In total, Limnofregata could be scored for 251 characters (54.1%) in the dataset.

To assess the monophyly and relationships of the Plotopteridae, four previously recognized members of this extinct family were included: Plotopterum joaquinensis (USNM 8927–cast of LACM 8927; [43]), Phocavis maritimus [60], Tonsala hildegardae (USNM 256518; [50]) and Copepteryx hexeris (Holotype: USNM 486682–cast of KMNH VP 200,006; Paratypes: USNM 243773–cast of KMNH VP 200,001; USNM 486684–cast of KMNH VP 200,002; USNM 243774–cast of NSMT VP 15035; [5]). Of the plotopterids included, only Phocavis was coded strictly from the literature [60]. Despite its fragmentary nature (known only from a tarsometatarsus) the inclusion of Phocavis in the current analysis is worthwhile, as this taxon has been described as the oldest, and possibly most basal member of Plotopteridae [5], [60], though Mayr [23] noted its overall similarity to Limnofregata and also raised the possibility that Phocavis represents the sister taxon to a Plotopteridae + Spheniscidae clade. The plotopterids included could be scored for the following proportions of characters in the dataset: Plotopterum joachinenesis, 3.1%; Phocavis maritimus, 7.5%; Tonsala hildegardae, 17.2%; Copepteryx hexeris, 44.2%. In addition to these four plotopterids, several other specimens of Plotopteridae were referred to for comparative purposes: Copepteryx titan (USNM 486685–cast of KMNH VP 200,004), Tonsala? sp. (USNM 243775–cast of KMNH VP 200,003).

In addition to Limnofregata and Plotopteridae, three other taxa purported to be stem members of extant pelecaniform families (or more inclusive clades) were included in the analysis. In constrast to Limnofregata and Plotopteridae, these taxa were coded exclusively from the primary and descriptive literature. Both Prophaethon shrubsolei from the lower Eocene London Clay of England [61], [62] and Lithoptila abdounensis from the upper Paleocene of Morocco [26], [53] are members of the extinct family Prophaethontidae, which has been recovered as the sister taxon to the extant Phaethontidae [26]. ?Borvocarbo stoeffelensis, a small, cormorant-like bird from the late Oligocene of Germany was recently described by Mayr [55], and includes a referred isolated foot that previously had been tentatively assigned to the extinct genus Oligocorax. Mayr [54], [63] noted the similarities between ?Borvocarbo stoeffelensis and extant ‘microcormorants’, but cautioned against referral of ?Borvocarbo stoeffelensis to crown or stem Phalacrocoracidae, noting that the species exhibits several plesiomorphies of Phalacrocoracoidea ( =  Anhingidae + Phalacrocoracidae). These three fossil taxa could be scored for the following proportion of characters in the dataset: Prophaethon shrubsolei, 20.5%; Lithoptila abdounensis, 24.4%; ?Borvocarbo stoeffelensis, 7.3%.

Character Sampling and ILD Tests

A total of 464 osteological characters were scored for each taxon (Appendix S2, Appendix S3). Characters can be divided into coarsely defined anatomical regions as follows: cranial skeleton, 95; axial skeleton, 11; pectoral skeleton, 188; pelvic skeleton, 169; miscellaneous, 1. 88 (19%) characters are new or have been formulated for phylogenetic analysis for the first time. The remaining characters, or some variation thereof, have been utilized previously in phylogenetic analyses. Characters were assembled from a variety of studies, with the primary sources being [3], [7], [23], [25], [26], [33], [64]. With regard to previously utilized characters, in general these were sampled according to the following criteria: 1) character states had to vary within the ingroup; 2) characters unique to only a single terminal taxon in the ingroup were not included; 3) characters and individual character states were independent of each other; 4) homology of the character and character states across the ingroup was clear and relatively uncontroversial; and 5) distinctions between character states were well-defined. Morphological traits were coded into binary or multistate characters. In cases where homology with a particular state in a taxon or set of taxa could no be confidently hypothesized for a character, these taxa were coded as inapplicable ( =  “−”) for that charcter. In the context of a maximum parsimony analysis this is effectively the same as treating these taxa as missing data ( =  “?”) for that charcter. Inapplicable characters were most problematic for the two outgroup taxa included, given their morphological dissimilarity to many members of the ingroup. However, only 16 of the 464 of the included characters, or 3.4%, required inapplicable codings in one or both outgroup taxa.

Four incongruence length difference (ILD) tests [65], were performed on the three primary anatomical partitions using the partition homogeneity test implemented in PAUP* 4.0b10 [66]. These analyses were performed with the fossil taxa excluded from the dataset, as several fossil taxa cannot be scored for characters in one or more anatomical partition (e.g., all included plotopterids lack cranial material), or can only be scored for a limited amount of characters (e.g., Prophaethon and Lithoptila are missing data for most pectoral characters). The first test compared all three major partitions (cranial, pectoral, pelvic) at the same time, utilizing heuristic searches of 500 pseudoreplicates with 25 random addition sequence replicates per pseudoreplicate. The remaining ILD tests were pairwise comparisons of anatomical partititions (i.e., cranial vs. pectoral, cranial vs. pelvic, pectoral vs. pelvic) and utilized heuristic searches of 200 pseudoreplicates with 20 random addition sequence replicates per pseudoreplicate.

Phylogenetic Analyses

Phylogenetic analyses of the taxon-character matrix were performed in PAUP* 4.0b10 [66]. Characters were equally weighted and treated as unordered. A heuristic search was performed with 10,000 random addition sequence replicates to obtain the most parsimonious trees for the dataset. Tree bisection and reconnection (TBR) was utilized as the branch-swapping algorithm for the heuristic search. Zero length branches were collapsed if they lacked support under any of the most parsimonious reconstructions, following ‘rule 1’ of Coddington and Scharff [67]. Clade support was quantified through bootstrap analysis [68]. Heuristic searches were performed on 2,000 pseudoreplicate datasets, with 10 random addition sequence replicates for each bootstrap search. The maximum number of trees saved for each random addition sequence replicate was set to 100 to prevent searches from becoming stuck on a large island of MPTs during any particular random addition sequence replicate. Though this search strategy reduces the amount of tree space explored for any given random addition sequence replicate, it allows for a much larger number of bootstrap replicates to be performed. Bremer support values were also calculated for each node in the strict consensus of all MPTs using TreeRot.v2c [69].

In addition to this analysis of the full dataset, a phylogenetic analysis was also performed with the eight fossil taxa removed. Three additional parsiomony analyses were performed analyzing each of the three major anatomical partitions (cranial, pectoral, pelvic). These analyses were also performed with the eight fossil taxa removed, as many of the fossil taxa included cannot be scored for characters in some partitions (e.g., plotopterids and cranial characters), or can only be scored for a small amount of characters (e.g., Prophaethon/Lithoptila and pectoral characters). These parsimony analyses were performed following the same protocol and methods described for the full dataset analysis listed above. Clade support was assessed using bootstrap analyses and Bremer support analyses as described above.

Several constraint analyses were also performed on the full dataset. Two of these assessed the relative support for recent higher-level phylogenenetic relationships of waterbirds [4], [29]. These two analyses involved setting up a backbone constraint tree that matched the topology recovered by either: 1) Livezey and Zusi's [4] morphological anlaysis, or 2) Hackett et al.'s [29] molecular analysis (Figure 1). Four additional analyses were performed that focus on the relationships of one or more members of the Pelecaniformes that have been contentious (see Introduction above). These included analyses that constrained: 1) the monophyly of a traditional Pelecaniformes; 2) the monophyly of a Plotopteridae + Spheniscidae clade; 3) the monophyly of a Balaeniceps + Scopus + Pelecanus clade; and 4) the monophyly of an Anhingidae + Sulidae clade. These constraint analyses were performed following the same protocol described for the primary phylogenetic analysis of the full dataset above.

At present, there are essentially no methods for assessing whether topologies alternative to the optimal tree/s are statistically significant worse fits to the character data in a parsimony framework. Non-parametric paired sites tests such as the Templeton test (a variation on the Wilcoxon signed ranks test; [70], [71]), the winning-sites test [72], and the Kishino and Hasegawa, or KH test [73] have been used extensively by morphological systematists, primarily because they can be implemented in a parsimony framework, and are included in popular phylogenetics software (e.g., PAUP* 4.0b10 [66]). However, all of these tests assume a null hypothesis where the expected difference in optimality score between alternative phylogenies is zero [74]–[76]. This requires that the topologies being compared must be specified a priori, and without reference to the data being used for the test. However, nearly all uses of these tests involve comparing alternative topologies to the optimal topology estimated from the data. This application guarantees that the null expectation of difference will always be larger than zero, and violates any assumption of a normal distribution of differences in optimality scores between topologies [76]. More recently, non-parametric tests, such as the Shimodaira-Hasegawa, or SH test [75] and the Approximately Unbiased, or AU test [77]; and parametric tests, such as the SOWH test [75], [76], have been developed that explicitly avoid these shortcomings. However, there is currently no implementation available for these tests using morphological data in a parsimony framework.

As noted by Goldman et al. [76], there is one possible modification of the KH test that allows for much more limited, but statistically valid, interpretation of its results in the context of the SH test. The P-value that would be obtained under the SH test is necessarily greater than or equal to half the P-value obtained by the KH test [76]. Thus, if the adjusted p/2 value from a KH test is greater than 0.05 (i.e., for a 5% significance level), which would indicate the inability to reject an alternate topology, than the P-value from the SH test would necessarily give the same conclusion [76]. However, in all cases where a KH test would indicate rejction of the alternative topology, (i.e., where KH test p/2<0.025), it is impossible to know whether the SH test would, or would not, indicate rejection of the alternate topology at a 5% significance leve (i.e., the SH test P-value will exceed the KH test p/2 value by an unknown amount) [76]. This severly limits the informativeness of the KH test, and essentially renders it an asymmetrical test of alternative topologies. If the KH test p/2>0.05, it can be concluded (on the basis of the SH test), that the present dataset cannot significantly reject the alternative topology as an equally good approximation of the phylogeny. Note however, that an additional confounding factor is that these statistical tests do not make corrections for assessing multiple trees, nor is a simple multiple-test correction such as the Bonferroni correction applicable to the problem [78].

KH tests were performed in PAUP* 4.0b10 [66] to assess differences between most parsimonious trees resulting from the unconstrained analysis of the full dataset, and most parsimonious trees obtained under the six constraint analyses outlined above. P-values were halved and interpreted following the recommendations of Goldman et al. [76]. Winning-sites and Templeton tests were also computed in PAUP* 4.0b10 [66], purely for comparison with the results of the modified KH tests, bearing in mind the statistical invalidity of these tests as outlined above.

Higher-level Phylogeny of Waterbirds

Phylogenetic analysis of the full dataset resulted in the recovery of six most parsimonious trees (MPTs), the strict consensus of which is presented in Figure 2. Individual MPTs were 1222 steps, with consistency and retention indices of 0.441 and 0.852, respectively. MPTs differed only in the relationships among the four plotopterids, and the relationship of Ciconia relative to Ardeidae and Threskiornithidae. The monophyly of individual waterbird families (where more than one member was included in the analysis) are well supported in most cases (e.g., Spheniscidae, Phaethontidae, Ardeidae, Threskiornithidae, Fregatidae, Sulidae, Anhingidae, Phalacrocoracidae).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Strict consensus of 6 MPTs from the full dataset analysis.

Tree length: 1222, C.I.: 0.441, R.I.: 0.852. Bootstrap proportions greater than 50% are shown below nodes. Bremer decay values greater than one are shown above nodes. Fossil taxa are indicated with a “†” superscript after taxon name. Branch colors are as in Figure 1. Waterbird families are indicated to the right of taxa, with members of a traditional Pelecaniformes indicated in black, and all others in grey.

https://doi.org/10.1371/journal.pone.0013354.g002

A monophyletic Pelecaniformes is not recovered, with Phaethontidae (tropicbirds) and the extinct Prophaethontidae comprising the sister taxon to Procellariiformes as part of a larger group including loons, grebes, and penguins (Figure 2). All other pelecaniforms are recovered in a monophyletic Steganopodes [79] (i.e., pelicans, frigatebirds, sulids, anhingas, cormorants). A sister taxon relationship between Anhingidae and Phalacrocoracidae is strongly supported, and this clade is recovered as the sister taxon to the extinct Plotopteridae. This Phalacrocoracoidea + Plotopteridae clade is recovered as the sister taxon to Sulidae in a monophyletic Suloidea. Fregatidae and Pelecanidae form successive sister taxa to Suloidea in a monophyletic Steganopodes (Figure 2). Balaeniceps and Scopus are recovered as successive sister taxa to Stegnopodes.

A ‘reduced’ Ciconiiformes clade (Ciconiimorphae sensu Livezey and Zusi [4]) is recovered as the sister taxon to the Scopus + Balaeniceps + Steganopodes clade. Monophyly of this ‘reduced’ Ciconiiformes clade is not strongly supported in the present analysis (Figure 2). The monophyly of both Ardeidae and Threskiornithidae is strongly supported, though their relationships to each other and to Ciconia are not clear. In three of the MPTs Ardeidae and Threskiornithidae are sister taxa, and Ciconia is recovered as the sister taxon to this larger clade. In the other three MPTs, Ciconia is recovered as the sister taxon to Threskiornithidae, and this larger clade is sister taxon to Ardeidae.

A large, basally diverging clade including loons, grebes, penguins, procellariforms and tropicbirds is recovered, similar to Livezey and Zusi's [4] Subdivision Pygopodo-tubinares, with the exception of the inclusion of tropicbirds (Figure 2). The monophyly of this larger clade is not particularly well supported, however, and none of the three basal-most divergences in this clade are supported by bootstrap values greater than 50% (Figure 2). As noted above, Phaethontidae and the extinct Prophaethontidae are recovered as the sister taxon to Procellariiformes. Penguins are recovered as the sister taxon to a well supported loon/grebe clade, and this larger clade forms the sister taxon to the procellariform/tropicbird clade (Figure 2).

Extant Taxa Only Analysis

Phylogenetic analysis of the extant taxa only dataset resulted in the recovery of seven MPTs (Figure 3). Individual most parsimonious trees were 1154 steps, with consistency and retention indices of 0.461 and 0.860, respectively. The MPTs are nearly identical to those recovered in analysis of the full dataset, with several notable exceptions. First, there is not unequivocal support for the monophyly of a large clade including loons, grebes, penguins, procellariforms, and tropicbirds. The monophyly of this clade is recovered in only six out of the seven MPTs, though two monophyletic subclades (one consisting of loons, grebes and penguins, and one consisting of procellariforms and tropicbirds) are recovered in all MPTs. In one MPT, the procellariforms + tropicbirds subclade and the loons + grebes + penguins subclade are recovered as successive sister taxa to all other ingroup taxa. Second, the large ‘reduced’ Ciconiiformes clade from the full dataset analysis is not recovered as monophyletic in all MPTs. In two MPTs, herons, threskiornithids, Ciconia, and Phoenicopterus are recovered as four separate lineages forming successive sister taxa to the Scopus + Balaeniceps + Steganopodes clade. In one MPT, a monophyletic clade of herons, threskiornithids, and Ciconia is recovered as the sister taxa to the Scopus + Balaeniceps + Steganopodes clade, and Phoenicopterus is recovered more basally as the sister taxon to this more inclusive group. Finally, in a single MPT, Phoenicopterus is recovered as the sister taxon to a large clade including loons, grebes, penguins, Procellariiformes, and tropicbirds. The positions of Fregata and Pelecanus have also switched relative to the MPTs from the full dataset analysis. In the extant taxa only analysis, Pelecanus is recovered as the sister taxon to Suloidea, with Fregata as the sister taxon to this larger clade in a monophyletic Steganopodes.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Strict consensus of 7 MPTs from the analysis of extant taxa only.

Tree length: 1154, C.I.: 0.461, R.I.: 0.860. Bootstrap proportions greater than 50% are shown below nodes. Bremer decay values greater than one are shown above nodes. Branch colors are as in Figure 1. Waterbird families are indicated to the right of taxa, with members of a traditional Pelecaniformes indicated in black, and all others in grey.

https://doi.org/10.1371/journal.pone.0013354.g003

ILD Tests and Partition Analyses

The incongruence length difference test comparing the three major anatomical regions (cranial, pectoral, pelvic) recovered a significant difference in phylogenetic signal between partitions (p = 0.002). The pairwise ILD tests suggest that this incongruence may primarily be between the pectoral and pelvic anatomical partitions, which was the only one of the three pairwise ILD tests that recovered significant incongruence (p = 0.005; cranial vs. pectoral p = 0.115; cranial vs. pelvic p = 0.205). The strict consenses of most parsimonious trees resulting from each of the partitioned analyses are presented in Figures 4–6.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. Strict consensus of 61 MPTs from the cranial partition analysis.

Tree length: 211, C.I.: 0.493, R.I.: 0.879. Bootstrap proportions greater than 50% are shown to the right of nodes. Bremer decay values greater than one are shown to the left of nodes. Branch colors are as in Figure 1.

https://doi.org/10.1371/journal.pone.0013354.g004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 5. Strict consensus of 7 MPTs from the pectoral partition analysis.

Tree length: 470, C.I.: 0.468, R.I.: 0.860. Bootstrap proportions greater than 50% are shown to the right of nodes. Bremer decay values greater than one are shown to the left of nodes. Branch colors are as in Figure 1.

https://doi.org/10.1371/journal.pone.0013354.g005

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 6. Strict consensus of 22 MPTs from the pelvic partition analysis.

Tree length: 388, C.I.: 0.503, R.I.: 0.879. Bootstrap proportions greater than 50% are shown to the right of nodes. Bremer decay values greater than one are shown to the left of nodes. Branch colors are as in Figure 1.

https://doi.org/10.1371/journal.pone.0013354.g006

Relationships within Steganopodes are relatively consistent between the partitioned analyses (Figures 4–6). A notable exception is the relative position of Papasula, which is recovered as the sister taxon to Sula in the pectoral partition analysis (as in the full dataset and extant taxa only analyses), and as the sister taxon to all other Sulidae in the cranial partition analysis. An additional interesting result is that neither the cranial nor pelvic partitioned analyses unambiguously resolve the relative positions of Fregata and Pelecanus to Suloidea, though the pectoral partition does recover a sister taxon relationship between Fregata and Suloidea. In light of the results of the extant taxa only analyses, which recover a sister clade relationship between Pelecanus and Suloidea, this suggests that the phylogenetic signal supporting a Pelecanus + Suloidea clade is cumulative across the three major anatomical partitions [80], as none of the remaining 12 characters (i.e., those not included in the partitioned analyses) provide unambiguous support for a closer relative relationship of Fregata or Pelecanus to Suloidea.

Relationships of the members of the ‘reduced’ Ciconiiformes clade differ considerably between the three partitioned analyses (Figures 4–6). In the MPTs from the cranial partition analysis, these taxa are recovered in a weakly supported (bootstrap support <50%) monophyletic clade, which also includes Scopus (Figure 4). In contrast, the MPTs from the pectoral partition analysis recover a sister taxon relationship between Ciconia and a Scopus + Balaeniceps + Steganopodes clade, with Threskiornithidae and Ardeidae in an unresolved polytomy basal to this group. Phoenicopterus is recovered as a member of the large basal clade including loons, grebes, penguins, procellariforms, and tropicbirds, though none of these splits are supported by bootstrap values greater than 50% (Figure 5). Finally, in the pelvic partition analysis, a monophyletic group consisting of the ‘reduced’ Ciconiiformes taxa is recovered, and Phoenicopterus and Ciconia are resolved as sister taxa closely related to Threskiornthidae (Figure 6). Additionally, Scopus and Balaeniceps are recovered as successive sister taxa to this ‘reduced’ Ciconiiformes clade, and this larger group is resolved as being one of the basal two phylogenetic splits in the ingroup (Figure 6).

The relative relationships of members of the large basal clade including loons, grebes, penguins, procellariforms, and tropicbirds (Figures 2, 3), also differ between the partitioned analyses (Figures 4–6). In the MPTs from the cranial partition analysis, this group is not monophyletic, with tropicbirds recovered as more closely related to the other ingroup taxa. A monophyletic clade including loons, grebes, penguins, and procellariforms is still recovered as one of the two basal splits in the ingroup, though within this group loons and grebes are not unambiguously monophyletic, and a sister taxon relationship between penguins and procellariforms is weakly supported (Figure 4). The MPTs of the pectoral partition analysis bear the most resemblance to the results of the full dataset and extant taxa only analyses (Figure 5). In these a large, basally splitting clade including loons, grebes, penguins, procellariforms, and tropicbirds is recovered, though Phoenicopterus is also recovered in this group. A clade including loons, grebes and penguins is weakly supported, as well as a sister taxon relationship between this group and Procellariiformes. In the pelvic partition analysis, tropicbirds are recovered as sister taxon to Steganopodes in a traditional Pelecaniformes (Figure 6). A monophyletic clade including loons, grebes, penguins and procellariforms is recovered as the sister taxon to Pelecaniformes, with an expanded Ciconiiformes clade more distantly related, as noted above. The monophyly of loons and grebes is strongly supported in the pelvic partition analysis, and this clade is weakly supported as the sister taxon to Procellariiformes. Penguins form the sister clade to this larger group (Figure 5).

Constraint Analyses

The constraint analysis enforcing the higher-level waterbird topology recovered by Livezey and Zusi's [4] morphological anlaysis resulted in the recovery of three MPTs, each of which were 1243 steps (21 steps longer than the unconstrained MPTs), with consistency and retention indices of 0.434 and 0.847, respectively. The constraint analysis enforcing the higher-level waterbird topology recovered by Hackett et al.'s [29] molecular analysis recovered three MPTs, each of which were 1371 steps (149 steps longer than the unconstrained MPTs), with consistency and retention indices of 0.393 and 0.820, respectively. The sister taxa relationships of the unconstrained fossil taxa in the MPTs of the constraint analyses are not different from in the unconstrained analyses. ?Borvocarbo stoeffelensis is still recovered as the sister taxon to Phalacrocorax; Plotopteridae is still recovered as the monophyletic sister taxon to Phalacrocoracoidea, Limnofregata is still recovered as the sister taxon to Fregata, and Prophaethon and Lithoptila are still recovered as successive sister taxa to Phaethon.

Constraint analyses enforcing a monophyletic traditional Pelecaniformes resulted in the recovery of six MPTs, (C.I. 0.438; R.I. 0.850). These MPTs are only nine steps longer than the MPTs of the unconstrained full dataset analysis (1231 vs. 1222 steps). As above, the relative sister taxon relationships of the fossil taxa in this constraint analyses are the same as in the full dataset analysis. In the MPTs, a Limnofregata + Fregata clade is recovered as sister taxon to Suloidea, with Pelecanus forming the sister taxon to this larger group. A monophyletic Phaethontidae + Prophaethontidae clade is recovered as the sister taxon to Steganopodes. Balaeniceps and Scopus are recovered as successive sister taxa to Pelecaniformes. Interestingly, enforcing a monophyletic Pelecaniformes also results in Phoenicopterus being recovered as the sister taxon to the large basal clade including loons, grebes, penguins, and procellariforms. Within this group, penguins and procellariforms are recovered as sister taxa, and this group is sister taxon to a monophyletic loon + grebe clade. The remaining members of the ‘reduced’ Ciconiiformes clade from the full dataset analysis are recovered in a monophyletic clade, with Ciconia forming the sister taxon to Threskiornithidae.

Constraint analyses enforcing the monophyly of a Plotopteridae + Spheniscidae clade resulted in the recovery of 12 MPTs (C.I. 0.440; R.I. 0.851). These MPTs are only four steps longer than the MPTs of the unconstrained full dataset analysis (1226 vs. 1222 steps). In the MPTs, a monophyletic Plotopteridae + Spheniscidae clade is recovered as the sister taxon to a clade including loons and grebes. The remaining relationships in the MPTs are nearly identical to those recovered in the full dataset analysis, with two exceptions. First, the relative relationships of the Limnofregata/Fregata clade and Pelecanus to each other are not resolved. Second, two alternate placements of Papasula: as the sister taxon to Sula, or as the sister taxon to all other sulids; are equally parsimonious.

Enforcing a monophyletic clade consisting of Balaeniceps, Scopus, and Pelecanus results in the recovery of 28 MPTs (C.I. 0.432; R.I. 0.846). These MPTs are 25 steps longer than the MPTs from the unconstrained full dataset analysis (1247 vs. 1222 steps). As above, the relative sister taxon relationships of the fossil taxa in this constraint analysis are the same as in the full dataset analysis. Much of the higher-level relationships outside of the Fregata + Suloidea clade are not resolved, with a basal polytomy including: a monophyletic clade of loons, grebes and penguins; Phoenicopterus; Ardeidae; Threskiornithidae; Ciconia; Procellariiformes; a monophyletic clade of Scopus, Balaeniceps and Pelecanus; and a monophyletic clade of Prophaethontidae and Phaethon. An Adams consensus of the 28 MPTs reveals that much of this poor resolution can be attributed to the uncertain phylogenetic placement of Phoenicopterus.

Constraints analyses enforcing the monophyly of an Anhingidae + Sulidae clade resulted in the recovery of 23 MPTs (C.I. 0.427; R.I. 0.843). MPTs are 39 steps longer than those from the unconstrained full dataset analysis (1261 vs. 1222 steps). Relationships recovered are virtually identical to those present in the MPTs from the unconstrained analysis, with the exception of several areas of less resolution within Suloidea: 1) the relationships within Morus; 2) the relationship of Papasula relative to Morus and Sula; and 3) the relationships within the ‘microcormorants’.

The results of the winning-sites, Templeton, and modified KH tests assessing the alternative constrained topologies are presented in Table 1. Note that for each set of constrained MPTs, only the P-value for the best fitting MPT is reported in order to make assessment of significance conservative. The p-values from the three paired-sites tests are all congruent in the rank order that the six suboptimal constrained topologies are placed in, which is also congruent with their rank order ascertained by the number of extra character changes implied by these topologies relative to the optimal unconstrained MPTs (Table 1). Topologies enforcing monophyly of Pelecaniformes, and monophyly of a penguin/plotopterid clade, represent the only two constrained topologies with KH test p/2 values >0.05, supporting the interpretation that these topologies would not be considered significantly worse fits to the present dataset than the optimal topologies under an SH test [76]. MPTs recovered under constraints matching the Livezey and Zusi [4] topology approach this threshold, with KH test p/2 values slightly below 0.05 (Table 1). However, a result where SH tests fail to reject an alternative topology (e.g., an inability to reject the null hypothesis) is difficult to interpret as positive evidence that the alternative tree is as good a fit to the character data as the optimal tree, as the SH test is known to be particularly conservative [76]–[78], [81]. Given these issues with interpretation, as well as the problems inherent in these paired sites tests as discussed above (see also [76], [78]), the results of these tests will not be discussed further.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Results of the constrained analyses and pairwise tests of topologies.

https://doi.org/10.1371/journal.pone.0013354.t001

Relationships of Prophaethontidae and Phaethontidae

In the present analysis, Prophaethontidae is recovered as a paraphyletic grade leading to Phaethontidae, with Prophaethon more closely related to modern tropicbirds than to Lithoptila, a result slightly different from that of Bourdon et al. [26], who recovered a monophyletic Prophaethontidae. Only two unambiguous synapomorphies support a Prophaethon + Phaethontidae clade to the exclusion of Lithoptila (42:0–>1; 59:0–>1), both of which were discussed by Bourdon et al. [26]. A larger clade of Prophaethontidae and Phaethontidae is recovered as the sister taxon to Procellariiformes, similar to the results recovered by Bourdon [25] and Bourdon et al. [26]. This Procellariiformes + (Prophaethontidae + Phaethontidae) clade is supported by 15 unambiguous synapomorphies (57:0–>1; 102:0–>1; 114:1–>2; 138:0–>1; 150:0->1; 152:0–>1; 188:0–>1; 248:0–>2; 259:0–>1; 320:0–>1; 339:0–>1; 379:0–>1; 399:0–>1; 461:0–>1; 463:0–>1), three of which exhibit no homoplasy on the MPTs (57:0–>1; 399:0–>1; 463:0–>1). Several of these characters are briefly discussed below.

29. Quadrate, orientation of the squamosal and otic condyles relative to the long axis of the skull: obliquely oriented, angle between 20-75° (0); nearly perpendicular, angle between 75-90° (1); nearly parallel, angle less than 20° (2) ([64]: character 148).

In most waterbird taxa, the squamosal and otic condyles of the quadrate are obliquely oriented relative to the long axis of the skull. Anhinga is unique in that the squamosal condyle is shifted markedly forward, such that the angle formed by the long axis of the skull and a transverse line through the center of both condyles is extremely acute, and the condyles are nearly parallel to the skull's long axis. However, in Prophaethontidae, Phaethontidae, Cochlearius, and contra Livezey and Zusi ([64]: character 148), Balaeniceps, the quadrate condyles are oriented much more strictly mediolaterally to each other, and nearly perpendicular to the long axis of the skull (Figure 7B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 7. Pelecaniform skulls in ventral aspect.

Pelecanus occidentalis FMNH 342303 (A), Phaethon aethereus FMNH 348136 (B), Sula sula FMNH 339372 (C), Fregata magnificens FMNH 339418 (D), and Limnofregata azygosternon FMNH PA755 (D, E). Scale bars equal 10 mm. Abbreviations: dtr, dorsal tympanic recess; md, mandible; pa, palatine; para, paroccipital process; pil, pila otica; pt, pterygoid; orc, orbital ( =  squamosal) cotyle; otc, otic cotyle; q, quadrate; quad/jug, quadratojugal/jugal.

https://doi.org/10.1371/journal.pone.0013354.g007

68. Dorsal tympanic recess, greatly enlarged, much longer than broad, extending rostral to and between cotylae quadratica in a figure-8 shape: absent (0); present (1) ([26]: character 32).

In Phaethon, Prophaethon and Lithoptila, the dorsal tympanic recess is greatly enlarged, and extends rostral to, and caudally between, the quadrate cotyles (Figure 7B). In ventral aspect, the outline of the dorsal tympanic recess in these taxa also takes on a slightly medially bent, figure-8 shape, with circular caudal and rostral portions, and a constriction at the rostral borders of the quadrate cotyles (Figure 7B). The rostral portion of the recess is slightly larger and more extensive medially.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 8. Pelecaniform right tarsometatarsi in cranial aspect.

Phaethon aethereus FMNH 348136 (A), Sula sula FMNH 339372 (B), Phalacrocorax africanus FMNH 368742 (C), and Phalacrocorax carbo FMNH 339390 (D). Scale bars equal 10 mm. Abbreviations: ehl, furrow for the tendon of musculus extensor hallucis longus; ex, sulcus extensorius; fov, fovea at proximal end of trochlea metatarsi III; int, eminentia intercondylaris.

https://doi.org/10.1371/journal.pone.0013354.g008

Monophyly of Steganopodes

A monophyletic Steganopodes consisting of Pelecanidae, Fregatidae, Sulidae, Anhingidae, Phalacrocoracidae, and the extinct Plotopteridae, is recovered in the present analysis. Steganopodes is supported by 28 unambiguous synapomorphies (11:0–>1; 44:0–>1; 80:1–>2; 83:0–>1; 135:0–>1; 140:0–>1; 160:0–>1; 169:0–>1; 263:0–>1; 268:0–>1; 274:2–>1; 280:0–>1; 285:0–>1; 287:0–>1; 300:0–>1; 312:0–>1; 315:0–>1; 356:0–>1; 374:0–>1; 376:0–>1; 381:0–>1; 390:1–>0; 395:0–>1; 397:0–>2; 398:0–>1; 404:2–>0; 440:0–>1; 448:0–>1), though only one exhibits no homoplasy across the MPTs (374:0–>1). Several of these characters are discussed in more detail below.

440. Tarsometatarsus, concave incisure in the medial side of the distal edge of trochlea metatarsal II: absent (0); present (1) ([64]: character 2351).

Among waterbirds, Pelecanus, Fregata, and Suloidea are unique in the presence of a distinct notch on the medial side of trochlea metatarsal II (Figure 9). This incisure is often best viewed in distal aspect, and is more pronounced in Suloidea than in Fregata and Pelecanus. Contra Livezey and Zusi ([64]: character 2351), this notch is not present in Phaethon, though this taxon does have a slightly expanded and medially everted plantar process on trochlea metatarsal II. Interestingly, the derived feature is present in Copepterxy hexeris (Figure 9D; [5]: Figure 10C), but it is not clear whether it is present in Phocavis ([60]: Figure 2F).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 9. Waterbird right tarsometatarsi in distal aspect.

Phaethon aethereus FMNH 348136 (A), Phoebastria nigripes FMNH 339601 (B), Eudyptula minor FMNH 106505 (C), Copepteryx hexeris USNM 243773–cast of KMNH VP 200,001 (D), Pelecanus erythrorhynchos FMNH 445082 (E), Sula sula FMNH 339372 (F), and Fregata minor FMNH 339421 (G). Scale bars equal 5 mm for (A–C, E–G), and 10 mm for (D). Abbreviations: in II, incisure in the medial side of the distal edge of trochlea metatarsal II; sul II, sulcus on the dorsal face of trochlea metatarsal II.

https://doi.org/10.1371/journal.pone.0013354.g009

Monophyly and Relationships of the Plotopteridae

Despite being highly incomplete in some cases (e.g., Phocavis, Plotopterum), the four purported plotopterids are recovered together in a monophyletic clade that is supported by eight unambiguous synapopmorphies (176:0–>1; 178:0–>1; 409:0–>1; 410:0–>1; 414:0–>1; 427:0–>1; 435:0–>1; 445:3–>2), three of which exhibit no homoplasy on the MPTs (178:0–>1; 409:0–>1; 435:0–>1). Several of these characters are described in more detail below. A variety of additional characters in the highly modified forelimbs of plotopterids also variably support the monophyly of the Plotopteridae (since many of these elements are unknown in Phocavis and Plotopterum, their status as synapomorphies are partially dependant upon method of character optimization). Many of these characters (e.g., characters 107, 159, 164, 165, 198, 210, 220, 229, 248, 256) are also present in penguins, interpreted on the MPTs recovered as being derived independently in the two clades.

156. Scapula, shape of acromial process: blunt to rectangular process (0); extremely compressed dorsoventrally, elongate, finger-like morphology (1). New Character.

The acromial process of the plotopterids Tonsala hildegardae (USNM 256518; [50]: Figure 4A) and Copepteryx hexeris (USNM 486682–cast of KMNH VP 200,006; [5]) is distinct not only for being extremely craniocaudally elongate, but the process is also strongly dorsoventrally compressed, giving the acromion a finger-like morphology in medial or lateral aspect (Figure 10). The elongate acromial process of these two plotopterids is also distinctly concave laterally (Figure 10A). Interestingly, scapulae referred to the late Eocene stem-Cariamidae taxon Elaphrocnemus phasianus also possess an extremely elongate acromial process that is somewhat dorsoventrally compressed, though not to the degree seen in plotopterids ([84]: Figure 1). The acromial process of Elaphrocnemus phasianus is also not distinctly concave laterally as in Plotopteridae ([84]: Figure 1). Mayr and Mourer-Chauviré [84] suggest that an elongate acromial process may function in strengthening the triosseal canal and acrocoracoid process of the coracoid.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 10. Pectoral girdle elements of several waterbird taxa.

Right scapula of Tonsala hildegardae USNM 256518 (A, B), Anhinga anhinga FMNH 339387 (C), and Eudyptula minor FMNH 106505 (D), in dorsal (A), and lateral (B–D) aspects. Left coracoid of Tonsala hildegardae USNM 256518 (E, H), Anhinga anhinga FMNH 339387 (F, I), and Pygoscelis antarctica FMNH 390994 (G, J), in lateral (E–G), and dorsal (H–J), aspects. Partial skeleton of Copepteryx hexeris USNM 486682–cast of KMNH VP 200,006 (K, L) in dorsal aspect. Scale bars equal 25 mm (A–D), 10 mm (E–J), and 25 mm (K, L). Abbreviations: acm, acromial process; acr, acrocoracoid process; apx, cranial apex of sternal keel; cf, coracoid facet; cor, coracoid; fur, furcula; hu, humeral articular facet; nt, medial notch to furcular facet; pro, procoracoid process; vert, dorsal vertebrae; rib, dorsal rib; sf, scapular facet; sca, scapula; str, sternum.

https://doi.org/10.1371/journal.pone.0013354.g010

Note that as the derived state is unknown in Plotopterum and Phocavis, this character state is only reconstructed as a synapomorphy of Plotopteridae under ACCTRAN character optimization in trees where Copepteryx and Tonsala are sister taxa.

Relationships of ?Borvocarbo stoeffelensis and Phalacrocoracidae

The late Oligocene ?Borvocarbo stoeffelensis [54], [55] is recovered as the sister taxon to Phalacrocoracidae in the present analysis. Three unambiguous synapomorphies support the monophyly of a ?Borvocarbo + Phalacrocoracidae clade (91:0–>1; 446:1–>2; 464:0–>1), one of which (464) exhibits no homoplasy in the MPTs. These characters are described in more detail below.

Monophyly of Suloidea

A monophyletic Suloidea (superfamily Suloidea sensu [3]; also considered parvorder Sulida [4], or suborder Sulae [5]) is well supported in the present analysis (Figure 2). This clade is supported by 20 unambiguous synapomorphies (2:1–>0; 8:1–>2; 22:0–>1; 35:0–>1; 41:0–>1; 47:0–>1; 63:0–>1; 64:0–>1; 66:0–>1; 71:0–>1; 87:0–>1; 96:0–>1; 112:2–>0; 158:1–>0; 181:2–>0; 358:0–>1; 367:3–>1; 373:1–>0; 458:0–>1; 461:0–>1), six of which exhibit no homoplasy on the MPTs (22:0–>1; 35:0–>1; 41:0–>1; 71:0–>1; 96:0–>1; 458:0–>1). Several of these characters are described in more detail below.

Relationships within the Sulidae

A monophyletic Sulidae is well supported in the present analyses (Figures 2,3). Sulidae is also strongly supported as the sister taxon to a clade comprised of Phalacrocoracoidea and Plotopteridae. Support for Morus is robust, with this genus being diagnosed by 16 unambiguous synapomorphies (3:0–>1; 48:0–>2; 108:0–>2; 113:0–>1; 152:1–>2; 177:0–>1; 186:0–>1; 193:0–>1; 235:0–>1; 236:0–>1; 245:1–>0; 251:0–>1; 363:0–>1; 365:0–>1; 383:0–>1; 408:1–>2), nine of which exhibit no homoplasy on the MPTs (3:0–>1; 48:0–>2; 108:0–>2; 177:0–>1; 193:0–>1; 235:0–>1; 236:0–>1; 363:0–>1; 383:0–>1). A single synapomorphy unites a clade of Morus bassanus and Morus capensis in the present dataset (297:1–>0; the absence of pneumaticity associated with caudomedial [i.e., contra-articular] face of the antitrochanter), though this character is quite labile across the MPTs (C.I. = 0.143). This grouping is identical to that tentatively proposed by Nelson [91] (see also Warheit [7], who was unable to resolve the relationships within Morus), though it contrasts with the sister taxon relationship of Morus capensis and Morus serrator recovered by Friesen and Anderson [6] in their analysis of partial cytochrome b sequences of sulids. The genus Sula is also strongly supported, diagnosed by 10 unambiguous synapomorphies (5:0–>1; 48:0–>1; 87:1–>2; 180:0–>1; 201:0–>1; 221:0–>1; 281:0–>1; 335:0–>1; 346:0–>1; 424:0–>1), 4 of which exhibit no homoplasy on the MPTs (48:0–>1; 201:0–>1; 221:0–>1; 346:0–>1). Relationships within Sula are well resolved, and congruent with both previous analyses of morphological data [6], [7], and partial sequences of the mitochondrial gene cytochrome b [6].

Papasula abbotti is resolved as the sister taxon to Sula in the present analysis (Figure 2), identical to the topology recovered by Warheit [7], but at odds with the sister taxon relationship with Morus recovered by Friesen and Anderson [6]. However, the position of Papasula is not particularly well supported, and the MPTs of the pelvic anatomical partition analysis support the placement of Papasula as sister taxon to all other Sulidae (Figure 6). Six unambiguous synapomorphies support a sister group relationship between Papasula and Sula (112:0–>1; 122:1–>0; 190:0–>1; 192:0–>1; 339:0–>1; 430:0–>1), only one of which exhibits no homoplasy on the MPTs (192:0–>1). Several of these characters are discussed below.

Relationships of Limnofregata

Limnofregata is robustly supported as the sister taxon to a monophyletic Fregata in the present analysis (Figure 2). 18 character states are reconstructed as unambiguous synapomorphies of the Limnofregata/Fregata clade in the set of most parsimonious trees (30:0–>1; 101:2–>1; 102:0–>1; 103:1–>0; 108:0–>1; 114:1–>2; 133:1–>2; 134:1–>0; 155:1–>0; 201:0–>2; 206:2–>1; 221:0–>2; 223:0–>2; 293:0–>1; 316:2–>1; 394:0–>1; 404:0–>1; 408:1–>2). Of these, 6 synapomorphies exhibit no homoplasy within the dataset (30:0–>1; 108:0–>1; 201:0–>2; 221:0–>2; 293:0–>1; 394:0–>1). Several of these characters are described in more detail below.

Character 30. Quadrate, shape of otic head in dorsal aspect: round or bulbous (0); compressed anteroposteriorly and distinctly elongate mediolaterally (1). New Character.

Most waterbirds have a quadrate otic head that is rounded or subspherical in dorsal aspect, and is not significantly broader mediolaterally than long anteroposteriorly. An anteroposteriorly compressed and mediolaterally elongate otic head is present in Fregata. A specimen of Limnofregata azygosternon (FMNH PA 720) includes a well-preserved right quadrate that is exposed in posterodorsal and medial view (Figure 11). The otic process is clearly visible, and possesses the mediolaterally elongate morphology present in Fregata.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 11. Right quadrates of several waterbird taxa.

Phalacrocorax carbo FMNH 339390 (A, B), Fregata minor FMNH 339421 (C, D, F), Phoebastria nigripes FMNH 339601 (E), and Limnofregata azygosternon FMNH PA 719 (G), in medial (A, D); dorsal (B, C); and medial/posterodorsal (E–G) aspects. In (B, C) the articular surface of the otic process of the quadrate is highlighted in magenta. Scale bars equal 5 mm. Abbreviations: op, otic process.

https://doi.org/10.1371/journal.pone.0013354.g011

108. Sternum, shape and relative craniocaudal length to mediolateral width of dorsal surface of sternal body: rectangular, sternal body longer than wide (0); square-shaped, sternal body wider than long (1); elongate rectangular, sternal body more than twice as long than wide (2) modified from ([64]: characters 1099, 1100, 1101).

Most waterbird taxa have a main sternal body that is longer anteroposteriorly than wide mediolaterally. The three extant species of Morus are unique in having an extremely elongate sternal body that is rectangular-shaped. Fregata is unique in possessing a sternum that is abbreviated craniocaudally, such that the anteroposterior length of the sternal body is less than its mediolateral width, and also less than the overall length of the coracoid. The holotype specimen of Limnofregata azygosternon includes a sternum that is preserved in left ventrolateral aspect, and is clearly abbreviated craniocaudally as in Fregata ([35]: Figures 2, 16, 17). An additional specimen of Limnofregata azygosternon (FMNH PA 755) preserves a sternum exposed in dorsal aspect, with the right coracoid in articulation and several fragmentary portions of ribs lying on the dorsal surface (Figure 12). As in USNM 22753, the sternum of FMNH PA 755 is box-like and mediolaterally wider than long anteroposteriorly. Olson ([35]: p. 2) first recognized the box-like and mediolaterally wide morphology of the sternal body as a similarity between Limnofregata and Fregata, and included this character in his diagnosis of the family Fregatidae, which included both taxa.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 12. Sterna of several waterbird taxa.

Ciconia abdimii FMNH 368771 (A), Pelecanus erythrorhynchos FMNH 445082 (B), Sula sula FMNH 339372 (C), Morus serrator FMNH 339366 (D), Fregata minor FMNH 339421 (E), and Limnofregata azygosternon FMNH PA 755 (F), in dorsal (A–E), and dorsal and slightly left lateral (F) aspects. Individual elements are color-coded in (F). Scale bars equal 25 mm. Abbreviations: clp, craniolateral process; co, coracoid; fu, furcula; hu, humerus; ri, ribs; st, sternum.

https://doi.org/10.1371/journal.pone.0013354.g012

206. Humerus, relative development and shape of deltopectoral crest: slightly protruding, low and rounded (0); strongly protruding and triangular (1); extremely reduced (2) ([23]: character 39); see also ([64]: character 1374).

Both Fregata and Limnofregata (Figure 13) possess strongly protruding deltopectoral crests that are triangular in outline in dorsal or ventral aspect. This character state is convergently present in Phoebastria and Puffinus.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 13. Right humeri of several waterbird taxa.

Ciconia abdimii FMNH 368771 (A), Fregata magnificens FMNH 339418 (B, C), and Limnofregata azygosternon UWGM 6919 (D, E), in medial (A, B; with distal portion enlarged), anterior (C), and anteromedial (D, E; with distal portion enlarged, and color-coded in E). Scale bars equal 50 mm. Abbreviations: dp, deltopectoral crest; pec, tubercle for attachment of m. pectoralis superficialis, pars deep; pro, scar for attachment of m. pronator superficialis; tsv, tuberculum supracondylare ventrale.

https://doi.org/10.1371/journal.pone.0013354.g013

293. Manus, proximodistally elongate fenestra on the distal third of the blade of II-1: absent (0); present (1). New Character, though see ([35]: p. 24).

The caudal, blade-like portion of manual phalanx II-1 is typically excavated into two large, proximodistally aligned fossae on its dorsal surface. The excavation on phalanx II-1 created by these fossae can sometimes be extreme enough to make this portion of the phalanx translucent, as in Scopus. In most waterbirds however, the distal fossa is not associated with any fenestration of phalanx II-1. In Fregata, a large, irregular and proximodistally elongate fenestra perforates the distal portion of the blade of phalanx II-1. As originally noted by Olson ([35]: p. 24) a similar fenestra is also present in the holotype of Limnofregata azygosternon (USNM 22753). Though perforation in such a fragile and thin area of the skeleton might easily be attributed to taphonomic or preparation damage in a single fossil specimen, both right and left elements of USNM 22753 preserve the fenestra, as well as portions of its smooth, rounded rims. In addition, several referred specimens (FMNH PA 720, FMNH PA 731; FMNH PA 755) also preserve the distal fenestra (Figure 14). Although the fenestra is slightly smaller in these specimens than in USNM 22753 and modern Fregata, the fenestrae in all Limnofregata specimens are proximodistally elongate and slightly elliptical. The fact that seven different elements of phalanx II-1 of Limnofregata possess a small, proximodistally elongate fenestra on their distal blade argues against the perforation being artifactual.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 14. Manual phalanges II-1 of several waterbird taxa.

Ciconia abdimii FMNH 368771 (A), Sula sula FMNH 339372 (B), Fregata minor FMNH 339421 (C), Fregata magnificens FMNH 339418 (D), Limnofregata azygosternon FMNH PA 731 (G), and Limnofregata azygosternon USNM 22753 (H), in ventral aspect; and left phalanges II-1 of Fregata magnificens FMNH 339418 (right element reversed) (E), and Limnofregata azygosternon USNM 22753 (F), in dorsal aspect. Scale bars equal 10 mm. Abbreviations: fen, fenestra.

https://doi.org/10.1371/journal.pone.0013354.g014

394. Tibiotarsus, medial ridge of trochlea cartilaginis tibialis hypertrophied, robust, and mound-like: absent (0); present (1). New Character, though see also ([64]: character 2170).

In most waterbirds, the medial ridge of trochlea cartilaginis tibialis on the posterior face of the distal tibiotarsus is not particularly more robustly developed than the lateral ridge. In Fregata, however, the medial ridge is mediolaterally widened, mound-like, and much more condylar in appearance than the lateral ridge, particularly at the proximal end of trochlea cartiliaginis tibialis on the posterior tibiotarsus (Figure 15). Olson ([35]: p. 27) originally alluded to the condition in Limnofregata, stating: “In posterior view the crest of the internal condyle is better developed than in Sula or Phaethon and more closely resembles the condition in Fregata”. The holotype of Limnofregata azygosternon (USNM 22753) includes a complete left tibiotarsus. The proximal third was removed and is contained on a smaller block that includes the sternum and distal third of the left femur. The distal two-thirds of the left tibiotarsus remain on the main block, and is exposed in caudal aspect, with its distal end adjacent to the left pubis and posterior portion of the ilium. The posterior face of trochlea cartilaginis tibialis is clearly visible, and the medial ridge is more robustly developed than the lateral, and condylar in appearance, similar to the condition in modern Fregata.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 15. Left tibiae of several pelecaniforms in distal aspect.

Pelecanus erythrorhynchos FMNH 445082 (A), Phalacrocorax auritus FMNH 348372 (B), Sula sula FMNH 339372 (C), and Fregata magnificens FMNH 339418 (D). Scale bars equal 5 mm. Abbreviations: lc, lateral condyle; lr, lateral ridge of trochlea cartilaginis tibialis; mc, medial condyle; mr, medial ridge of trochlea cartilaginis tibialis.

https://doi.org/10.1371/journal.pone.0013354.g015

Relationships of the Limnofregata + Fregata Clade

A sister-taxon relationship between the Limnofregata/Fregata dyad and Suloidea is recovered in the present analysis, and supported by 16 unambiguous synapomorphies, two of which exhibit no homoplasy across the MPTs. Though recent studies of higher-level avian relationships based on molecular data have begun to converge on robust support for a Fregata + Suloidea clade [27], [29], this remains slightly incongruent with morphological phylogenetic analyses. Nearly every morphological phylogenetic analysis of pelecaniform relationships recovers Pelecanus as sister-taxon to Suloidea, with Fregata as the sister-taxon to this larger clade [3], [4], [26]. Notable exceptions to this pattern include the analyses of Mayr [22], which could not resolve the relative position of Fregata and Pelecanus to Suloidea (though Mayr [22] utilized the traditional topology for his discussion of character optimization); and Mayr [23], which recovered Fregata as closer to Suloidea than Pelecanus, though a clade formed of Spheniscidae and the extinct Plotopteridae were recovered as the actual sister-taxon to Suloidea. While the ‘switching’ of positions of Pelecanus and Fregata recovered in this analysis may seem like a relatively minor detail, the fact that the phylogenetic position of Pelecanus in the waterbird tree is so contentious, and that these two alternate positions for Fregata render the node being temporally calibrated by Limnofregata either congruent or incongruent with molecular topologies, merits further discussion of the characters supporting the (Limnofregata/Fregata) + Suloidea clade in the present analysis.

Pelecaniform Fossil Calibration Points and Implications for Temporal Patterns of Diversification

Both Prophaethon [95], and more recently Lithoptila [27], [28], have been utilized as fossil calibration points for analyses of divergence timing in higher-level avian clades. Of these two taxa, Lithoptila represents the oldest record of Prophaethontidae and stem-group tropicbirds, with the oldest specimens recovered from upper Paleocene (Thanetian) sediments [26], [53], though Prophaethon, from the lower Eocene (Ypresian) London Clay, is only slightly younger [61]. The only previously published phylogenetic analysis to include members of Prophaethontidae [26] also recovered strong support for the recovery of both Prophaethon and Lithoptila as stem tropicbirds. Thus, at present, the use of either of these taxa as a calibration point for the node uniting Phaethontidae with its closest extant relative appears to be well justified.

However, considerable disagreement exists regarding the closest extant relative of Phaethontidae. Recent analyses of morphological data have suggested that Phaethontidae are sister taxon to Procellariiformes [25], [26], (the present study), or to the remaining members of a traditional Pelecaniformes [4]. Several recent higher-level molecular studies sampling nuclear genes suggest that Phaethontidae is a member of an ecologically diverse clade of neoavians variably referred to as “Metaves” [19], [27], [29]. However, the relative relationships of tropicbirds within this clade are not particularly well supported or congruent across these studies. Additionally, a recent mitogenomic study of higher-level avian relationships recovered Spheniscidae as the sister taxon to Phaethontidae, with this larger clade as the sister taxon to a (Fregatidae + Sulidae + Phalacrocoracidae) clade [28]. Accordingly, though the phylogenetic placement of Prophaethon and Lithoptila, and their use as a calibration point for the divergence of tropicbirds from their closest extant relative may be well-supported, the relative impact this calibration point has on patterns of temporal divergence in higher-level avian phylogeny will be dependent on the successive extant sister group relationships of modern tropicbirds, which remain extremely labile at present.

Given the topology recovered in the present analysis (Figure 2), a late Paleocene (Thanetian) age for Lithoptila implies that the lineages leading to Procellariiformes and to the loon + grebe + penguin clade had both diverged prior to the Eocene. This is largely consistent with the fossil record of the Procellariiformes. The oldest fossil material tentatively assigned to this clade is the nearly complete right humerus of Tytthostonyx glauconiticus from the Late Cretaceous or early Paleocene of New Jersey [96]. However, Olson and Parris ([96]: p. 16) stressed the tentativeness of a referral to Procellariformes, noting similarities between Tytthostonyx and Charadriformes and Pelecaniformes. Similarly, Bourdon et al. ([53]: p. 759) noted similarities between Tytthostonyx and Lithoptila, as well as Charadriformes, and rejected procellariform affinities for this taxon. As De Pietri et al. ([97]: p. 1) note, fragmentary Eocene fossils from throughout the world have been referred to extant procellariform families [98]–[100]; and more complete stem procellariforms are known from the early Oligocene [97], [101]. The future analysis of this material in a rigorous phylogenetic framework would allow more definitive statements regarding the earliest records of stem and crown Procellariiformes.

Similarly, the oldest purported members of the Gaviiformes are known from the Cretaceous, and include Neogaeornis wetzeli, an isolated tarsometatarsus from Campanian/Maastrichtian sediments of Chile [102]; and Polarornis gregorii, a partially associated skeleton from the Late Cretaceous of Antarctica [103]. However, doubts have been raised regarding the gaviiform affinities of both of these taxa [104], [105]. The late Eocene Colymboides anglicus [106], [107], consisting of a coracoid (and subsequently a referred humerus and frontal portion of a skull; [108]) constitutes the next oldest fossil material referred to Gaviiformes, though more informative, partially associated skeletons are not known until the early Oligocene [104], [109].

The oldest stem penguin fossils include several partial skeletons from the Paleocene of Antarctica [110] and New Zealand [111]. Of these, Waimanu manneringi represents the oldest known stem penguin, dated with calcareous nannofossils as late early Paleocene (60.5–61.6 Myr; [111]: p. 1145, suppl. mat.). The relationships of Waimanu manneringi, and its slightly younger sister taxon Waimanu tuatahi, have also been evaluated in several phylogenetic analyses of morphological, and combined molecular and morphological datasets, and are consistently recovered as the basal-most members of Sphenisciformes [85], [111], [112]. Given that relatively complete, associated skeletons of Waimanu have been recovered, the temporal data associated with this fossil material is well constrained, and the phylogenetic relationships of Waimanu have been rigorously tested, this material represents a key calibration point for use in studies of higher-level avian diversification.

Limnofregata has likewise been utilized as a fossil calibration point for analyses of divergence timing in recent studies [27], [28]. Though Limnofregata has long been considered as closely related to extant frigatebirds [23], [35], [36], [51], [59], [113], the present study represents the first time this taxon has been evaluated in a modern phylogenetic analysis. Limnofregata is robustly supported as the sister taxon to modern frigatebirds in the present analysis (Figure 2). The characters supporting this clade are distributed throughout the entire avian skeleton, which argues against a close relationship being erroneously inferred due to ecomorphological convergence in a particular anatomical subregion [114]. In addition, many of the characters supporting a Limnofregata/Fregata clade are preserved in multiple specimens of Limnofregata, which provides confidence that character states have been accurately assessed, and cannot be attributed to taphononomic or preparation damage of an individual specimen (e.g., the distal fenestration of manual phalanx II-1). This robust phylogenetic placement, coupled with the detailed age control on the the early Eocene Fossil Butte Member of the Green River Formation (from which the majority of Limnofregata specimens are known), which recent radiometric dating indicates is 51.97+/−0.16 Myr [58], makes Limnofregata particularly attractive as a potential fossil calibration point for future studies of higher-level avian divergence timing.

The present study also represents some of the first support from morphological phylogenetic analysis for a sister taxon relationship between Fregatidae and Suloidea. Additional evidence supporting the monophyly of a Fregata + Suloidea clade includes the unique synovial intraramal joints in the mandibles of these taxa ([115]; [64]: character 712; though see also Louchart et al. [116], who consider the joint to be present in Pelecanus). Most previous analyses of morphological data recover Pelecanus, and then Fregata, as successive sister-taxa to Suloidea [3], [4], [26]. Morphological support for a sister taxon relationship between Fregatidae and Suloidea is noteworthy, as a growing body of evidence from molecular phylogentic analyses of higher-level avian relationships also recovers this clade, often with strong support [19], [27]–[29]. Thus, there is somewhat of a provisional consensus across morphological and molecular studies on the extant sister group of Fregatidae, and accordingly the node being calibrated by Limnofregata fossils. An important caveat however, is that relative waterbird relationships outside of this Fregata + Suloidea clade still remain largely incongruent between morphological and molecular analyses (Figure 1), and recent studies [4], [19], [23], [26]–[29] often do not recover strong statistical support for these higher-level clades. Thus, the relative impact that a temporal calibration from Limnofregata would provide will be highly topology dependant, and involves which clades represent successive sister taxa to the Fregata + Suloidea clade.

Given the topology recovered in the present analysis (Figure 2), Limnofregata represents the earliest representative of Steganopodes, and suggests that the lineages leading to Fregatidae and Suloidea diverged by the early Eocene. This is largely in agreement with the fossil record of Suloidea, particularly if plotopterids belong within this clade (see below). As previously noted, this implies an enormous gap in the fossil record of Fregatidae, which, with the exception of Limnofregata, includes only Quaternary fossils from oceanic islands [36], [59]. In addition, the age of Limnofregata and the paraphyletic arrangement of Scopus, Balaeniceps, and Pelecanus to the Fregata + Suloidea clade recovered in the present analyis (Figure 2), suggest that the lineages leading to the former three taxa also diverged by the early Eocene. This again implies a large amount of missing fossil record for these three groups, as the Scopidae includes only one fossil taxon, Scopus xenopus, known from a distal tarsometatarsus and partial coracoid from the early Pliocene of South Africa [36], [117]; the Balaenicipitidae is known from only a few more fossil occurrences [36], the oldest of which is Goliathia andrewsi from the early Oligocene Jebel Qatrani Formation in Egypt [118]; and the fossil record of Pelecanidae, while slightly more extensive than that of Scopidae or Balaenicipitidae, reliably extends only to the early Oligocene of France [36], [90], [116]. This anomaly could stem from many factors, including the possibility that these lineages were represented by species-poor clades throughout their histories (e.g., both Scopidae and Balaenicipitidae are monotypic families, and the approximately seven species of extant Pelecanus are generally considered to represent recent divergences), or that much of these lineages' biogeographic history took place in areas that have poor fossil records, or have not been well-sampled (e.g., the current distributions of Scopidae and Balaenicipitidae are much more restricted than many other, more cosmopolitan, waterbird families). Alternatively, the phylogenetic position of these taxa recovered in the current analysis may be incorrect. As noted earlier, a growing body of evidence from molecular studies supports the monophyly of a clade including Pelecanus, Balaeniceps, and Scopus [19], [20], [24], [27], [29], and divergence dates recovered from molecular clock studies utilizing these topologies are in closer accord with the known fossil records for these three lineages ([27]: Figure 2). The divergence times recovered by Ericson et al. ([27]: Figure ESM-9) for the split between Pelecanus and its closest relative varied between ∼30 Myr for their PATHd8 analysis and ∼50 Myr for their Penalized Likelihood analysis. The former of which is more closely in accord with the earliest fossil record of the Pelecanidae, a nearly complete skull and neck from Rupelian strata (28.25–33.00 Myr) of southeastern France [116].

The majority of plotopterid specimens described are from late Oligocene and Miocene deposits [43], [49], [50], [59]. The oldest member of the family is Phocavis maritimus, from the Late Eocene to Early Oligocene Keasey Formation of the Pacific northwestern United States [60], [119]. Establishing the exact age of the Keasey Formation has been problematic ([119]: p. 16474). When Goedert ([60]: p. 99) originally described Phocavis, which is from the upper part of the informal middle member of the Keasey Formation, he considered the Keasey Formation to be Late Eocene, correlated with the Priabonian Chronostratigraphic Stage (40–36.6 Myr; [120]). This correlation was followed by Warheit [37], though Warheit ([38]: Appendix 2.1) placed Phocavis in the middle Eocene, with no clear justification. However, Prothero and Hankins [119] subsequently revised the age of the Keasey Formation utilizing magnetic stratigraphy, and suggested that the middle portion of the Keasey Formation was correlative with Chron C13r (33.5–34.6 Myr), which spans the Eocene–Oligocene boundary. In addition to Phocavis, Olson and Hasegawa ([5]: p. 742) briefly mention “as many as six species” of plotopterids in Japan from “older beds that are probably early Oligocene (but possibly late Eocene)”, though these specimens remain to be described in detail.

Given the phylogenetic position of Plotopteridae (Figure 2), this suggests that the lineages leading to Sulidae and to Phalacrocoracoidea had diverged by the Eocene–Oligocene boundary. This is largely consistent with the fossil record of Sulidae, which includes possible stem members such as Eostega lebedinskyi, from the middle Eocene of Romania [52], [121], Masillastega rectirostris, from the middle Eocene of Messel, Germany [51] (though see also Mlíkovsky [52], who refers this species to Eostega), and Sula ronzoni from the early Oligocene of France [36]. However, the relationships of these taxa remain to be tested in a rigorous modern phylogenetic analysis. The earliest definitive record of Phalacrocoracoidea is not known until the late Oligocene (∼24.7 Myr), with the presence of the stem phalacrocoracid ?Borvocarbo stoeffelensis from the Enspel fossil site in Germany ([55]; Figure 2; see also below). Fossil members of the Anhingidae are not known until the early Miocene [122]. Potential older members of Phalacrocoracoidea are known, including undescribed material referred to Phalacrocoracidae from the Eocene–Oligocene Phosphorites du Quercy, France [36], [123], and a partial premaxilla referred to Phalacrocoracidae from the early Oligocene Jebel Qatrani Formation in Egypt [118]. In addition, the enigmatic Plotoplotus beauforti from the Paleocene (see [124]) of Sumatra has been considered to exhibit affinities with pelecaniforms, particularly Anhingidae [36], [40], [42]. Once again, the relationships of taxa representing these purported earlier occurrences remain to be rigorously evaluated in a phylogenetic framework. Also, given the uncertainty surrounding the precise phylogenetic placement of Plotopteridae [5], [23], [50] (Figure 2), and the fact that a sister taxon relationship between Plotopteridae and penguins (as advocated by Mayr [23]) is only slightly less parsimonious in the present dataset, extreme caution should be excerised in using any member of the Plotopteridae as a potential fossil calibration point for divergence time studies.

A partial right foot, tarsometatarsus and distal tibiotarsus referable to ?Borvocarbo stoeffelensis (this material was originally considered ?Oligocorax sp. by Mayr [54]; Mayr [55] later described a new, nearly complete specimen as ?Borvocarbo stoeffelensis and referred this material to the new taxon) has also been utilized in one analysis of avian divergence timing [27]. This specimen was employed as a calibration of stem Phalacrocoracidae, and hence, the Anhingidae/Phalacrocoracidae split. Recovered from the late Oligocene fossil site of Enspel, dated at roughly 24.7 Myr [55], [125], ?Borvocarbo stoeffelensis represents the most complete and oldest fossil that can be referred to stem Phalacrocoracidae, or stem Phalacrocoracoidea (Anhingidae + Phalacrocoracidae). However, the present study represents the first time ?Borvocarbo stoeffelensis has been included in a morphological phylogenetic analysis, where it is recovered as the sister taxon to all other included Phalacrocoracidae (Figure 2).

Mayr ([55]: p. 940) cautioned against a straightforward assignment of ?Borvocarbo stoeffelensis to stem Phalacrocoracidae, based on the presence of two characters interpreted as plesiomorphic relative to Phalacrocoracoidea: 1) paroccipital processes that are less pointed and caudally everted; and 2) a shorter bicipital crest that meets the humeral shaft at a steeper angle. The present analysis recovered the first character as a synapomorphy of Suloidea (character 66). However, due to the potential distortion of the rear of the skull of ?Borvocarbo stoeffelensis, this taxon was coded as “?”, or missing data, for this character in the present analysis. The second character listed by Mayr ([55]: p. 940) was not included in this analysis. However, two additional characters that Mayr ([55]: pp. 939–940) listed as potentially uniting ?Borvocarbo stoeffelensis and Phalacrocoracidae were also not included in the present analysis: 1) a well-developed crista nuchalis sagittalis along the midline of the skull, which is present in Phalacrocoracidae with the exception of the ‘microcormorants’ (see also [33]: character 3); and 2) an accessory transverse cranial crest present caudal to crista nuchalis transversa. Thus, it is likely that inclusion of these additional characters would only increase the support for the recovery of ?Borvocarbo stoeffelensis as a stem Phalacrocoracidae. Despite this, the treatment of ?Borvocarbo stoeffelensis as a fossil calibration point for stem Phalacrocoracidae should still be viewed with some caution. If, as suggested by Mayr [55], [63], many of the characters supporting a close relationship between ?Borvocarbo stoeffelensis and Phalacrocoracidae actually represent secondary reversals in Anhingidae, then a ‘cormorant-like’ morphology would be expected for fossil taxa spanning the stem of both Phalacrocoracoidea and Phalacrocoracidae.

Relationships of the Plotopteridae

The recovery of a monophyletic clade of extinct Plotopteridae as sister-taxon to the Phalacrocoracoidea suggests that wing-propelled diving evolved independently in Plotopterids and penguins, contra Mayr [23]. This study thus confirms previous work [43], [59], [50], which suggested that the skeletal anatomy of plotopterids represents a remarkable case of convergent evolution on the more familiar penguin body-plan. However, despite this exceptional degree of morphological convergence, the analysis of osteological characters in a modern phylogenetic framework is able to parse out the numerous similarities in the penguin and plotopterid pectoral girdle and appendages as the result of homoplasy (e.g., characters 107, 159, 164, 165, 198, 210, 220, 229, 248, 256). An important caveat however, is that members of Spheniscidae were relatively sparsely sampled for the current analysis (two taxa), and no fossil sphenisciforms, particularly along the stem leading to extant penguins, were included. As noted above, the character state distributions present in some fossil penguins may in some cases increase (e.g., character 272), or decrease (e.g., character 427), support for a close relationship between penguins and plotopterids. Furthermore, enforcing a monophyletic clade of Spheniscidae + Plotopteridae results in trees that are only four steps longer than the MPTs from the unconstrained analysis of the full dataset.

An additional pattern is present in the dataset that lends support to the phylogenetic position of plotopterids recovered in the current analysis. Although there are numerous characters (particularly in the pectoral girdle) shared by plototperids and penguins (see above discussion), there appear to be few, if any synapomorphies that diagnose the nodes subtending the penguin clade (i.e., those that include loons, grebes, procellariforms, and tropicbirds) that are present in plotopterids. On the contrary, there are character states present in plotopterids that represent synapomorphies of higher-level clades within Steganopodes. For example, three unambiguous synapomorphies of the Fregata + Suloidea clade (175:1–>2; 341:0–>1; 413:0–>1) are present in one or more plotopterid. Furthermore, three unambiguous synapomorphies that diagnose Steganopodes (169:0–>1; 395:0–>1; 440:0–>1) are also in one or more plotopterid. Three unambiguous synapomorphies diagnosing the Balaeniceps + Steganopodes clade (155:0–>1; 187:0–>1; 434:0–>1) are also present in one or more plotopterid. Finally, one unambiguous synapomorphy diagnosing the Scopus + Balaeniceps + Steganopodes clade (143:0–>1) is present in all three plotopterids for which this character can be scored. These synapomorphies are also distributed throughout the entire skeleton, and not concentrated in a particular anatomical region. If the phylogenetic position of plotopterids recovered in the present analysis (Figure 2) is correct, than this sort of pattern of hierarchically nested synapomorphies present in the skeletons of plotopterids is to be expected, while morphological convergence related to the evolution of wing-propelled diving would result in plotopterids sharing character states with penguins, but not with the larger clades (e.g., those that include loons and grebes, or loons, grebes, procellariforms, and tropicbirds) that penguins are nested within. Additional comparative anatomical work on fossil penguins and plotopterids, coupled with expanded taxon sampling and more taxon-specific character sampling will likely yield further insight into the phylogenetic relationships of Plotopteridae.

Partitioned Analyses and Implications for Patterns of Character Evolution

Pairwise ILD tests revealed that the pectoral and pelvic anatomical partitions might possess discrepant phylogenetic signals. However, these results should be interpreted with caution, as the ILD test is a rather coarse tool for examining dataset incongruence, and previous workers have noted the potential for increased type 2 error (falsely rejecting the null hypothesis of dataset congruence) when: many invariant and/or parsimony uninformative characters are included [126], [127]; the level of homoplasious characters varies between partitions [128]; the size of the partitions being analyzed varies greatly [129], [130]; or character partitions are evolving at different rates [130]. Thus the interpretation of detecting significant incongruence between the pectoral and pelvic partitions is not straightforward, and could be an artifact (e.g., related to the relative sizes of these partitions), or the result of other biologically interesting phenomena (e.g., the possibility of different levels of homoplasy or rates of evolution in disparate partitions).

The results of the partitioned phylogenetic analyses also reveal that much of the discrepancy between partitions is localized outside of Steganopodes, and can primarily be attributed to a few problematic taxa (e.g., flamingos, tropicbirds, hammerkop). Furthermore, the nodes that conflict between the MPTs of the partitioned analyses are for the most part not well supported in any of the individual partitioned analyses (Figures 4–6), or the analysis of the full dataset (Figure 2). In addition to documenting the phylogenetic lability of several taxa (e.g., flamingos, tropicbirds, hammerkop), these partitioned analyses also aid in revealing signals that are not at first apparent in the full dataset analysis (Figure 2). These include the fact that the phylogenetic signal supporting the traditional relationship of a Pelecanus + Suloidea clade to the exclusion of Fregata (in the extant taxa only analyses) appears to be the result of cumulative hidden support across the three major anatomical partitions [80]. An additional interesting pattern recovered from the partitioned phylogenetic analyses is that the strong sister taxon relationship between loons and grebes is supported primarily by characters of the pelvic girdle (Figures 4–6). Partitioned analyses also reveal that pelvic characters alone are not sufficient to recover a monophyletic Sulidae. Gaining a better understanding of the patterns of phylogenetic support present in different anatomical regions could potentially go a long way toward evaluating the confidence that can be placed in the phylogenetic affinities of more fragmentary fossil waterbird specimens.

Congruence and Conflict in Higher-level Waterbird Phylogeny

As noted in the introduction, higher-level avian phylogeny is far from robustly resolved, and the relationships of Pelecaniformes in particular remain contentious. Given this uncertainty, a brief discussion of the patterns of congruence and conflict between the present study and the most recent and exhaustive morphological [4] and molecular [29] analyses of higher-level waterbird phylogeny is warranted, as well as a consideration of several potential explanations for conflicting topologies.

Rigorously assessing patterns of congruence and conflict between datasets can be problematic in and of itself [131]. While agreement or disagreement in optimal topologies from different datasets can be taken as congruence or incongruence per se, these patterns can also be the result of idiosyncrancies in the datasets, due to differential taxon or character sampling, choice of outgroup/s, or even different methods of analysis [132]–[135]. Additionally, conflict between datasets can often be overestimated if the issue of clade support is not taken into consideration [135].

For the purposes of this discussion, bootstrap values ≥70% are treated as strong support for a clade in a given analysis. This is a pragmatic, but arbitrary cutoff, and there are obvious caveats to this criterion, including the facts that support values are not directly comparable across different analyses, bootstrap values for the studies discussed were calculated under different optimality criterion (maximum likelihood in [29] and maximum parsimony in [4] and the present study), and the bootstrap in general tends to be conservatively biased [78], [136]. Similar methods and thresholds have been used previously for discussing support in empirical studies. For example, Mayr et al. [137] utilized a cutoff of bootstrap support >60% in a comparison of phylogenetic datasets of ‘higher land birds’, Smith et al. [138] employed a threshold of >70% bootstrap support to identify strongly supported conflicting nodes in molecular and morphological estimates of echinoid phylogeny, and Wiens et al. [139] utilized bootstrap values ≥70% as their threshold for designating strong support in a combined analysis of snake phylogeny. Given this criterion for assessing strong support, “congruence” would represent cases where a particular clade is strongly supported in two or more different datasets, while “conflict” would represent cases where a taxon or clade is strongly supported in different positions in two or more datasets. Bootstrap support values were taken directly from Livezey and Zusi ([4]: Figure 14), and Hackett et al. ([29]: Figure 2). For the present study, bootstrap values from the analyis including extant taxa only (Figure 3) were utilized. When applicable, comparisons to other recent studies, as well as the additional analyses of the present dataset (Figures 2, 4–6), are drawn as well.

Employing this strategy for comparing the present dataset and those of Livezey and Zusi [4] and Hackett et al. [29] yields a variety of cases of congruence. To begin with, the monophyly of two relatively uncontested waterbird clades, Ardeidae and Procellariiformes, is strongly supported by all three datasets. However, only Livezey and Zusi [4] included a relatively diverse sampling of the Ardeidae, with five different genera sampled from a family of approximately 64 species. Both the present dataset and that of Hackett et al. [29] included only two herons, which allows a relatively limited test of group monophyly. Similarly, all three datasets recovered the clade Procellariiformes with 100% boostrap support, though they each sampled only five taxa from all four procellariform families. This represents a phylogenetically broad, yet sparse, sample of the approximately 110 living species of this avian order. The current analysis and that of Livezey and Zusi [4] are both congruent in recovering a sister taxon relationship between Phoebastria (“Diomedea” in [4]) and Puffinus, though this does not imply a sister taxon relationship between Diomedeidae and Procellariidae, as the present dataset only sampled one member of the each family, and Livezey and Zusi [4] recovered their two sampled members of Procellariidae, Puffinus and Pachyptila, as successive sister taxa to Phoebastria (“Diomedea” in [4]). Both the Livezey and Zusi [4] and Hackett et al. [29] analyses recover strong support for a sister taxon relationship between Sphenisciformes and Procellariiformes. This is in contrast to the results of the present analysis, which recover Sphenisciformes as the sister taxon to a loon/grebe clade, and Procellariiformes as sister taxon to tropicbirds, though neither relationship is strongly supported. Interestingly, when tropicbirds are excluded from the dataset, MPTs are recovered that include a sister taxon relationship between penguins and procellariforms, which is supported in 63% of boostrap replicates (using the same phylogenetic methodology and search strategies as given in the “Methods” section above). Thus, there is clearly some character support in the present dataset for the procellariform/penguin clade that is well supported in both the Livezey and Zusi [4] and Hackett et al. [29] analyses.

Three other areas of congruence between the datasets center around relationships within Pelecaniformes. All three analyses recover an Anhingidae/Phalacrocoracidae clade ( =  “Phalacrocoracoidea”) with 100% bootstrap support. As noted in the introduction, a close affinity between Anhingidae and Phalacrocoracidae has generally been accepted as uncontroversial, with the exception of several mitochondrial datasets that recovered an unconventional sister taxon relationship between Anhingidae and Sulidae [2], [17]. Kennedy et al. [17] have suggested that long-branch attraction, exacerbated by a short internal branch separating Sulidae, Anhingidae, and Phalacrocoracidae, is likely confounding these datasets. The Livezey and Zusi [4], Hackett et al. [29], and present analyses also all strongly support the monophyly of Suloidea, a clade including Sulidae, Anhingidae, and Phalacrocoracidae. This clade has been recovered in nearly every modern phylogenetic study that has tested pelecaniform and/or waterbird relationships, and there is virtually no controversy regarding its monophyly [22]. Finally, both the current analysis and that of Livezey and Zusi [4] are congruent in recovering strong support for a monophyletic Steganopodes. However, as discussed above, the relative relationships of Pelecanidae and Fregatidae to Suloidea differ between these two analyses, though are not in “strong” conflict with each other, applying the criteria used for this discussion.

Several significant incongruencies between the present analysis, Livezey and Zusi [4], and Hackett et al. [29] are apparent, most of which involve problematic taxa or clades that have been noted previously. Chief among these is the recovery of strong support for a loon/grebe clade in the present analysis and Livezey and Zusi [4], and strong support separating these taxa in Hackett et al. [29]. A loon/grebe clade has been repeatedly tested and rejected by molecular phylogenetic analyses (see review by Mayr [140]), with grebes strongly supported as the sister taxon to flamingos, and this clade recovered well outside of a monophyletic waterbird assemblage. Recently, a growing body of morphological characters have been identified that may also support a close relationship between flamingoes and grebes [29], [140]–[142]. Interestingly, Livezey and Zusi ([4]: p. 48) noted that the majority of characters supporting a loon/grebe clade are from the pelvic girdle and hindlimb. The results of the partitioned analyses from the present dataset also bear out this pattern. The pelvic partition analysis recovers 100% bootstrap support for a sister taxon relationship between loons and grebes, while both the skull and pectoral partition analyses recover these two lineages in polytomies with other taxa (Figures 4–6). Coupled with the fact that both loons and grebes are foot-propelled pursuit divers, this pattern raises the possibility that this relationship is the result of morphological convergence in the pelvis and hindlimb of these taxa.

Wiens et al. [143] outlined a list of three explicit criteria for detecting whether adaptive convergence has misled a phylogenetic study. The first, “strong morphological support for a clade that unites the taxa that share the similar selective environment” ([143]: p.502), appears to be satisfied for loons and grebes, particularly with reference to characters in the pelvic girdle and hindlimb. Wiens et al.'s ([143]: p. 503) second criterion, “Evidence that the characters that unite the putatively convergent clade are associated with the shared selective environment” is slightly more difficult to establish. Hinic-Frlog and Motani [144] reported correlations between foot-propelled diving and the narrowing and elongation of the pelvis, as well as the proximal expansion of the cnemial crest; and indeed, several of the morphological characters supporting a loon/grebe clade in the present analysis are related to these structures (e.g., characters 296, 316, 324, 360, 361). However, the association of these morphological characters with foot-propelled diving remains to be demonstrated across a broader sample of taxa, and with rigorous phylogenetic comparative methods. The third criterion of Wiens et al. ([143]: p. 503), “Phylogenetic evidence that the species that share the common selective environment are not actually a monophyletic group, preferably consisting of strong support for the contradictory clades from two or more unlinked molecular data sets”, does appear to be satisified. Thus, although the explanation of adaptive convergence related to foot-propelled diving remains a tempting explanation for the loon/grebe signal in morphological datasets, more rigorous functional and comparative phylogenetic analyses are needed to support this claim.

Another point of incongruence present between the present analysis, Livezey and Zusi [4], and Hackett et al. [29], is the position of Pelecanoides within Procellariiformes. Livezey and Zusi [4] recover this taxon as the basal-most split within the order, while the present analysis recovers Pelecanoides as the sister taxon to a monophyletic Hydrobatidae. Hackett et al. [29] recover Pelecanoides as even more highly nested within Procellariiformes, as the sister taxon to Puffinus. The postions of Pelecanoides in the partitioned analyses are somewhat labile, which may explain some of this conflict (Figures 4–6). As noted above, Procellariiformes are much more diverse than the five taxa that each of the three phylogenetic studies have sampled, so taxon sampling, particularly with respect to the diverse Procellariidae, may be playing a role in this incongruence. However, a more detailed study of relationships within Procellariiformes, albeit based soley on the mitochondrial cytochrome b gene, by Nunn and Stanley [145] recovered an arrangement of procellariform families that is generally congruent with that of Hackett et al. [29], with the exception of the position of Oceanitinae.

The relative positions of both Ciconia and Phoenicopterus are also strongly incongruent between the analyses of Livezey and Zusi [4] and Hackett et al. [29], though not necessarily between either of these datasets and the present analysis, which has difficultly confidently placing either of these taxa (Figures 2–6). The relationships of flamingos within higher-level avian phylogeny as inferred from morphological data have long been contentious, with purported affinities including Anseriformes, Ciconiidae, and Recurvirostridae [30], [142]. Livezey and Zusi [4] recover flamingos and storks as strongly supported sister taxa, well nested within a clade of other ciconiiforms. Flamingos and storks are particularly labile in the present analyses (Figures 2–6). The results of the partitioned analyses (Figures 4–6) suggest that the phylogenetic signal driving their union as sister taxa and derived placement relative to other ciconiiforms may be predominantly coming from characters in the hindlimb (e.g., characters 337, 384, 393, 415, 418), though few characters in the present dataset appear to be unique to these two taxa. As previously noted, a large body of recent molecular evidence supports a sister taxon relationship between grebes and flamingos, with this couplet placed outside the waterbird clade [140]. The position of storks in recent molecular studies of higher-level avian phylogeny has been more uncertain. Hackett et al. [29] recover storks with strong support as the basal-most split of a mixed clade of pelecaniforms and ciconiiforms, though Ericson et al. [27] failed to resolve the position of storks relative to other waterbirds, and the mitogenomic study of Brown et al. [28] recovered storks in an unorthodox position as the sister taxon to steatornithid caprimulgiforms in their optimal topology, though these authors noted that this arrangement was suspect.

Another strong point of conflict between the present analysis and the analyses of Livezey and Zusi [4] and Hackett et al. [29] concerns the relationships of Scopus, Balaeniceps, and Pelecanus within waterbirds. Livezey and Zusi [4] recover Pelecanus as the sister taxon to Suloidea, with frigatebirds and tropicbirds as successive sister taxa to this clade in a monophyletic Pelecaniformes. Balaeniceps is recovered as the well-supported sister taxon to Pelecaniformes, while Scopus is allied with several other ciconiiforms. The results of the present analysis are more similar to those of Mayr [22], [23] and Bourdon et al. [26], in recovering a paraphyletic grade including Scopus, Balaeniceps, and Pelecanus, leading to a Fregata + Suloidea clade (Figures 2,3). Though it should be noted that the position of Pelecanus and Fregata is variable between these studies, and these datasets included few [22], [23], or no [26], other members of a traditional Ciconiiformes. In contrast, a large body of molecular data, including the anlaysis of Hackett et al. [29], supports a monophyletic clade of Scopus, Balaeniceps, and Pelecanus, though the relationships among these three taxa are not robustly resolved (see review by Mayr [140]).

As originally noted by Cottam [146] (though see also Mayr [22], [140]), there are several features suggestive of a close relationship between Pelecanus and Balaeniceps. An additional character identified in the present study that is unique to Pelecanus and Balaeniceps is the presenc of a rugose tuberosity distal to the bicipital crest on the proximal humerus (character 205). Both Pelecanus and Balaeniceps also have sterni with a costal margin that is extremely long relative to other waterbirds (character 127). Scopus, Pelecanus, Balaeniceps, and also Fregata appear to be unique in possessing a impression for m. biceps brachii on the coracoid that is situated significantly cranially, well above the facies articularis clavicularis (character 181). Note however, that this likely represents a facies articularis clavicularis that is more caudally placed on the cranial end of the coracoid, rather than a cranial migration of impressio m. biceps brachii. The condition in Fregata is difficult to code considering the fusion of the coracoid and furcula. However, the impressio ligamenti acrocoracohumeralis is situated well cranial to the area of fusion between the furucla and coracoid, and a pit-like depression that likely represents impressio m. biceps brachii is located just ventrally and caudomedially to the cranial end of impressio ligamenti acrocoracohumeralis, and is also well cranial to the fused furcula and coracoid. The possibility thus exists that morphological characters supporting a Scopus, Pelecanus, and Balaeniceps have gone previously unnoticed.

A final prominent area of conflict between the present analysis and the analyses of Livezey and Zusi [4] and Hackett et al. [29] regards the relationships of Phaethontidae (tropicbirds). The analysis of Livezey and Zusi [4] recovers tropicbirds in a traditional position, as the well supported, basal-most split within a monophyletic Pelecaniformes. Tropicbirds are also fairly nested within waterbirds in this analysis, being subtended by two more inclusive, well-supported clades. In contrast, the present analysis recovers tropicbirds as the sister taxon to Procellariiformes, similar to recent studies by Bourdon [25] and Bourdon et al. [26]. This position is not particularly well supported in the present analyses (Figures 2–6), though the clade uniting Balaeniceps and Steganopodes (and thus excluding tropicbirds), is strongly supported, which conflicts with the results of Livezey and Zusi [4]. In contrast, a growing body of molecular data (e.g., [19], [27]), including the Hackett et al. [29] analysis, strongly supports the placement of tropicbirds well outside of Pelecaniformes, and even outside the waterbird clade, though the close relatives of tropicbirds remain variable in these studies. The recent mitogenomic study of Brown et al. [28] recovered tropicbirds as the sister taxon to penguins, nested within a larger clade primarily composed of waterbirds. These authors' noted the dubious nature of this novel placement however, as was similarly the case with the placement of storks in their analysis [28].

The phylogenetic postion of tropicbirds is thus strongly incongruent and considerably variable among both morphological and molecular studies of higher-level avain phylogeny. Indeed, Cracraft ([3]: p. 834) noted that with regard to the Pelecaniformes, “phaethontids are the most aberrant family of the order”. Given that the divergence of crown Phaethontidae is relatively recent, likely within the past several million years [31], and that stem tropicbirds constrain the split of Phaethontidae from its closest extant relative to more than 55 Myr [26], [53], it is clear that the phylogenetic placement of tropicbirds could be susceptible to long-branch attraction in molecular datasets. Similar circumstances are present for both grebes and flamingos, where long-branch attraction has also been considered as a possible confounding factor [4]. In the present analysis, a variety of characters supporting a close relationship between tropicbirds and Procellariiformes are listed and described above in the section “Relationships of Prophaethontidae and Phaethontidae”. In addition to these, there are several characters where states present in either of the stem tropicbird fossil taxa Prophaethon and Lithoptila are different from those present in extant tropicbirds, and add support for a possible affinity with Procellariiformes (e.g., characters 232, 239, 366), or help to reinforce that the states present in Phaethontidae are not synapomorphic with the same states present in the basal steganopods Pelecanus and Fregata (e.g., characters 226, 362). Though the sister taxon relationship between tropicbirds and Procellariiformes recovered in the present analysis and by Bourdon [25] and Bourdon et al. [26] should be regarded as tentative, it is intriguing that the mosaic of character states present in stem tropicbirds adds support to this hypothesis, a potential phylogenetic benefit of fossil data that has long been recognized [147]–[150].

With regard to the patterns of higher-level waterbird phylogeny recovered in the present analysis and those of Livezey and Zusi [4] and Hackett et al. [29], there generally is more agreement between the former two morphological datasets than between either of these and the latter molecular dataset. Similarly, more conflict seems to be present between the Hackett et al. [29] phylogeny and either the present phylogeny or that of Livezey and Zusi [4], than between the two morphological phylogenies. Neither of these patterns is perhaps surprising. Potential causes for these conflicts are numerous, and include: 1) taxon sampling artifacts (with regard to the position of Pelecanoides); 2) limited character sampling (with regard to possible affinities between Scopus, Balaeniceps, and Pelecanus); 3) adaptive convergence (with regard to support for a loon/grebe clade, but possibly also a stork/flamingo clade); and 4) long-branch attraction (with regard to the aberrant placements of tropicbirds, the mitochondrial support for an Anhinga/Sulidae clade, and possibly also support for a flamingo/grebe clade). Many of these potential causes remain speculative however, and require further evaluation with rigorous comparative phylogenetic methods. Finally, the possibility remains that many higher-level waterbird clades represent temporally deep divergences that are separated by relatively short internodes. This pattern is born out by recent analyses of molecular [27]–[29] and morphological [4], [93] data, which both appear to recover higher-level internodes that are shorter, in terms of inferred character changes, relative to shallower internodes and terminal branches subtending and comprising the tips of the phylogenies. Paleontological evidence supports such a pattern as well, with many modern orders in place and anatomically distinct by the Paleogene, and evidence of pre-Cenozoic records for most neoavian clades scarce and often controversial [26], [151].

Several patterns regarding higher-level relationships inferred from the present morphological dataset are also noteworthy. To begin with, the significant conflict and incongruence regarding the position of several taxa in recent molecular analyses (e.g., [27], [29]) and the Livezey and Zusi [4] morphological analysis, is reduced in the present dataset. These cases involve the phylogenetic positions of: 1) Ciconia and Phoenicopterus; 2) Scopus, Balaeniceps, and Pelecanus; and 3) Phaethontidae. In none of these three cases are the results of the present analysis congruent with recent molecular analyses (e.g., [27], [29]), but strongly conflicting signals in the morphological dataset appear to be absent. Furthermore, it is evident from the partitioned analyses (Figures 4–6) that incongruence may be attributed to characters in a particular anatomical region, rather than wholesale conflicting signal throughout the skeleton. This gives hope that problematic morphological characters capable of misleading phylogenetic analyses may be able to be identified using modern comparative methods, a point stressed previously by McCracken et al. [152]. The results of the present analysis also suggest that the inclusion of stem fossil members of highly modified and deep diverging extant lineages may aid in resolving the phylogenetic position of problematic taxa (e.g., tropicbirds), by preserving deep synapomorphies and revealing superficial convergence. Clearly however, there is need for more morphological and molecular data, more rigorous evaluation of this data, and more comprehensive phylogenetic analyses, a point that all avian systematists appear to be in agreement on [4], [27]–[29], [93], [150], [151]. Despite the obvious need for well-supported morphological phylogenies that can accurately place fossil taxa in order to assess the tempo of avian diversification, the continued collection and intergration of morphological and molecular datasets will ultimately provide reciprocal illumination of higher-level avian phylogeny and evolution.

Conclusions

The monophyly and phylogenetic relationships of the avian order Pelecaniformes were assessed through the analysis of a morphological phylogenetic dataset of waterbirds encompassing 59 taxa and 464 characters. Parsimony analyses do not recover a monophyletic Pelecaniformes, recovering tropicbirds as distantly related to the remaining members of the order, which are supported as a monophyletic Steganopodes (pelicans, frigatebirds, sulids, darters, cormorants). Relationships of extant pelecaniforms inferred from morphology are more congruent with molecular phylogenies than previously assumed, though notable conflicts (e.g., the positions of Pelecanus and Phaethon) remain. ILD tests suggest that some major anatomical partitions of the dataset may possess different phylogenetic signals, though partitioned phylogenetic analyses reveal that these discrepancies are localized outside of Steganopodes, and can primarily be attributed to a few problematic taxa (e.g., flamingos, tropicbirds, hammerkop), or poorly supported nodes. The Plotopteridae, an extinct family of wing-propelled divers, are recovered as the monophyletic sister taxon to a cormorant–darter clade, suggesting numerous convergent adaptations in the pectoral limbs of plotopterids and penguins. However, support for this topology is relatively weak, and the higher-level relationships of Plotopteridae recovered here remain tentative.

The relationships of several fossil pelecaniforms representing key calibration points for studies of higher-level avian diversification are well resolved in the present study. These include Limnofregata (sister taxon to Fregatidae), Prophaethon and Lithoptila (successive sister taxa to Phaethontidae), and ?Borvocarbo stoeffelensis (sister taxon to Phalacrocoracidae). The sister taxon relationships of these fossil taxa are robust to alternate phylogenetic hypotheses, and do not change when ‘backbone’ phylogenetic constraints based on recent morphological and molecular studies of higher-level avian phylogeny are imposed. However, the successive outgroup relationships of several of these “stem fossil + crown family” clades (e.g., Limnofregata/Fregatidae; Lithoptila/Phaethontidae) remain highly variable and poorly supported across recent studies of avian phylogeny. Thus, the impact these fossil calibrations have on future studies of higher-level avian temporal diversification will depend heavily on the extant sister taxon relationships of Phaethontidae and Fregatidae, as well as the increased resolution and support of deep nodes in avian phylogeny.