The Australo-Papuan (Sahul) region possesses a high degree of endemism in birds and other elements of its biota, due in part to its long isolation from other major landmasses. In some cases these may date to the breakup of the Gondwanan supercontinent and many millions of years of isolation. Subsequently, a complex combination of plate collisions and the accretion of many terranes with differing historical origins have contributed to the unique biotic assemblage found in the region today (Hall 2001). More recently, the region was invaded by organisms from elsewhere, particularly southeast Asia (e.g., Schodde and Mason 1999). Nevertheless, a high degree of regional endemism remains, particularly noticeable given the short distance to Indo-Malayan influences, and is reflected in biogeographic generalizations such as Wallace's Line (Brown and Lomolino 1998).

In many instances, however, it is not clear whether organisms in the Australian region are closely related and the product of local divergence, representing endemic radiations, or are the result of multiple independent invasions from the north and west. Resolving such cases requires phylogenetic analyses of the groups to determine whether the endemics of the Australian region comprise monophyletic clades.

One long-standing enigma has been the relationships of six genera of hawks, comprising eight species (e.g., Dickinson and Remsen 2013), restricted to the Australo-Papuan region. These include the monotypic genera Hamirostra, Harpyopsis, Lophoictinia, and Megatriorchis, together with Erythrotriorchis and Henicopernis, both of which contain two species. Although traditional treatments, such as that of Peters (1931), placed these genera in four diverse subfamilies within the Accipitridae, Amadon (1978) regarded their relationships as problematic. He pointed out, for example, that Erythrotriorchis and Megatriorchis had tail:wing ratios inconsistent with those of typical accipiters, and that although Lophoictinia looked like a milvine kite, Hamirostra had buteonine proportions. Nevertheless, Stresemann and Amadon (1979) adopted a sequence similar to that used by Peters (1931), and subsumed Erythrotriorchis and Megatriorchis into Accipiter, placed Harpyopsis with the large harpy (Harpia, Morphnus) and monkey-eating (Pithecophaga) eagles, put Henicopernis adjacent to the Eurasian honey-buzzards (Pernis), and placed both Hamirostra and Lophoictinia with the milvine kites. Later, Amadon and Bull (1988) published a similar classification. However, these assemblages were not supported by any phylogenetic analyses, and first Olsen et al. (1993), and later Schodde (1993), suggested that the six genera might comprise representatives of an ancient Australian raptor fauna. Subsequently, Thiollay (1994) and Olsen (1995) postulated that the six genera actually might represent a single clade, possibly allied to Accipiter: a regional radiation of diverse accipitrids.

Gamauf and Haring (2004) investigated the relationships of Pernis and Henicopernis, but, using only 382 bp of the rapidly evolving mitochondrial cyt-b gene, they were unable to ascertain the relationships between the two. Griffiths et al. (2007) used DNA sequences from a nuclear protein-coding gene, RAG-1, to investigate the relationships among many genera of the Accipitridae. However, because those authors limited their survey to species for which freshly preserved tissue samples were available, they were only able to examine the relationships of three of these Austral endemic genera. They reported that Harpyopsis was close to the great or harpy eagles, as Amadon and Bull (1988) had suggested, but Hamirostra and Lophoictinia formed a clade restricted to Australia. Lerner and Mindell (2005) found the same relationships of these three genera using DNA sequences from two mitochondrial genes and a nuclear intron; they also did not have access to the remaining three genera and five species. Thus, were Erythrotriorchis, Henicopernis, and Megatriorchis found to be related to Hamirostra and Lophoictinia using DNA, there would be evidence for a significant radiation of raptors restricted to the Australia-New Guinea region, as suggested by Olsen (1995) on the basis of egg morphology.

Here we report on our investigation of the phylogenetic relationships of all the genera and species of Australia and New Guinea endemic hawks, using slowly evolving nuclear DNA sequences from the RAG-1 exon. This gene previously has proved useful for phylogenetic inference in hawks (Griffiths et al. 2007), falcons (Griffiths et al. 2004), and other non-passerine birds (Groth and Barrowclough 1999); more generally, it has provided the basis for our current classification of all passerine birds (Barker et al. 2002, Barker et al. 2004).

Methods

RAG-1 sequences of Harpyopsis novaeguineae from Papua-New Guinea and Hamirostra melanosternon and Lophoictinia isura from Australia were reported by Griffiths et al. (2007). We were able to obtain recent tissue samples of Henicopernis longicauda and Megatriorchis doriae from Papua-New Guinea; tissue samples of Henicopernis infuscatus and the two species of Erythrotriorchis were not available, so we extracted DNA from toe pads from specimens of these species in the collections of the American Museum of Natural History, the Bishop Museum, and the Western Australian Museum. In Griffiths et al. (2007), a single species of Circus (C. aeruginosus) had been sequenced; for this study, we sequenced two additional species (C. assimilis and C. cyaneus) to improve our understanding of the relationship between Accipiter and Circus. The source and voucher information for the new taxa used in this study are provided in the Appendix.

DNA was extracted using standard procedures; previously described RAG-1 primers were used to amplify and sequence the exon from the fresh tissues (Groth and Barrowclough 1999). Because the DNA from the toe pads was old, fragmented, and of poor quality, a novel set of 33 closely spaced RAG-1 primers (available from the authors) were designed to target the 5′ end of the gene from hawks. Previously obtained accipitrid RAG-1 sequences were used for this purpose (Griffiths et al. 2007). The labeled sequenced fragments of DNA were run out on an ABI 3730xl DNA Analyzer; contigs were assembled and aligned by eye using Sequencher software.

We examined the sequences for potential quality problems by checking for unexpected stop codons, indels not a multiple of three base pairs, and sequences with more than two chromatographic peaks at any single site. We checked sequences for contamination by foreign PCR product by examining the divergence between each pair of sequences using a sliding 300 bp window; unexpectedly similar or abnormally divergent pairs of sequences were flagged for further examination. We also checked for contamination by foreign PCR product by examining the distribution of heterozygosity among taxa for outliers. We examined the sequences for base composition heterogeneity among taxa by comparing the proportion of third codon positions that were either cytosine or guanine to those that were adenine or thymine.

The new sequences were added to the prior (Griffiths et al. 2007) RAG-1 dataset; this included sequences from four outgroups: Turkey Vulture (Cathartes aura), Prairie Falcon (Falco mexicanus), Gray Parrot (Psittacus erithacus), and Tawny Owl (Strix aluco). We inferred phylogenetic trees using likelihood methods using PAUP* version 4.0b10 (Swofford et al. 1996, Swofford 2001); for the initial analyses, all three amino acid codon positions were treated as a single data partition. Because several of the new sequences were obtained from toe pads and consequently were shorter than the sequences used in Griffiths et al. (2007), we treated nonsequenced base pairs, as well as gaps, as missing; multistate sites within taxa were treated as uninformative. We used the AIC criteria in Modeltest version 3.06 (Posada and Crandall 1998) to choose an appropriate initial likelihood model and set of parameter estimates for the evolution of the sequences. After an initial TBR heuristic search, starting with a neighbor-joining tree, we reestimated the likelihood parameters on the resulting maximum-likelihood tree and began a subsequent search for a new maximum-likelihood tree consistent with the reestimated parameters; this process was iterated until the resulting tree was consistent with the starting parameters (Sullivan et al. 2005). We used a χ² test to determine whether the maximum likelihood tree for these hawks was consistent with a clock-like model of DNA substitution (Huelsenbeck and Crandall 1997, Posada and Crandall 1998). We conducted a likelihood bootstrap procedure (1000 nonparametric replicates) using RAxML version 7.3.2 (Stamatakis 2006).

We conducted several additional phylogenetic analyses to confirm the standard maximum likelihood results; first, we performed a second maximum likelihood bootstrap (1000 replicates) analysis using RAxML in which first, second, and third codon positions were treated as separate partitions, each with its own model of substitution (three partition analyses). Second, we performed a parsimony bootstrap (100 replicates) analysis using PAUP* with five random additions of taxa and SPR branch-swapping within each replicate. Finally, we performed a Bayesian reconstruction using the program MrBAYES version 3.2.1 (Ronquist et al. 2012) for 10 million generations with 25% burn-in and chain-sampling every 1000 generations. A preliminary assessment of the sensitivity of the analysis to several branch length priors was conducted (Marshall et al. 2006). Convergence of chains was assessed using comparisons of log-likelihood scores and cumulative posterior probabilities.

We used a standard permutation procedure (SOWH-test) with a likelihood ratio test to determine whether the maximum likelihood topology obtained with our data was statistically superior to the likelihood of the hypothesis that all eight species in the six endemic genera formed a monophyletic clade (Hillis et al. 1996, Swofford et al. 1996). We compared the likelihood score of the best unconstrained tree given our data to that of the score of the best tree obtained subject to the constraint that all the members of the six endemic genera formed a clade. We generated 100 simulated parametric datasets using parameters obtained from the best constrained tree using SEQ-GEN vers. 1.3.2 (Rambaut and Grassly 1997). We obtained a null distribution of likelihood differences between constrained and unconstrained scores by comparing the differences between the likelihood of the best constrained tree and that of the best tree obtained in a new, unconstrained, TBR heuristic search for each of the 100 simulated datasets (Goldman et al. 2000).

Results

We obtained 2872 bp of RAG-1 sequence from the fresh tissues of Megatriorchis doriae, Circus assimilis, and C. cyaneus, and 2869 bp from Henicopernis longicauda; from the toe pad samples we obtained 714 bp from Erythrotriorchis buergersi and E. radiatus, and 711 bp from Henicopernis infuscatus. The sequences were easily aligned with the prior accipitrid RAG-1 sequences; a single three basepair deletion was required to align the two species of Henicopernis with the other samples. That single codon deletion coincided in position with a deletion previously described for Hamirostra and Lophoictinia. The initial (5′) basepair for all the new sequences corresponded to position 84 of the published chicken (Gallus gallus) RAG-1 sequence (GenBank M58530: Carlson et al. 1991). The new sequences have been deposited in GenBank; the accession numbers are provided in the Appendix.

The new sequences had base and amino acid compositions consistent with prior accipitrid RAG-1 sequences. No stop codons occurred in the aligned sequences and none of the sequences was unexpectedly similar to or divergent from any other. The patterns of heterozygosity provided no suggestion of contamination in the new sequences, including those obtained from toe pad samples; the maximum heterozygosity among the new samples was 0.003, for E. radiatus.

The AIC criterion in Modeltest suggested that GTR + Γ + I (general time reversal base pair substitutions with Γ-distributed evolutionary rates plus invariant sites) was the appropriate model for nucleotide substitution for this exon. We fixed the AIC estimated parameters for an initial heuristic likelihood search. The tree resulting from this first search was then used to reestimate the parameters for the model and start a second search. A tree with a superior likelihood score was not found; thus, the tree was consistent with its parameters (Sullivan et al. 2005). We computed the likelihood of this tree with and without the assumption of a molecular clock; clock-like substitution rates were rejected (χ²  =  225.4, df  =  70, P < 0.005).

The phylogenetic tree based on this single likelihood partition is shown in Fig. 1. It is quite similar to the tree reported by Griffiths et al. (2007) and is largely bifurcating with only a few multifurcations. All phylogenetic analyses placed the Australo-Papuan endemic genera in four positions on the tree; First, Hamirostra, Lophoictinia, and the two species of Henicopernis formed a clade embedded within the pernine kites; however, within the pernines, they were not sister to the honey-buzzard (Pernis). These four species also shared the only (3 bp) deletion found among the accipitrid sequences. Second, Harpyopsis was sister to the large Neotropical harpy eagles (Harpia and Morphnus). Third, the two species of Erythrotriorchis were sister to each other and in turn sister to several accipiters. Finally, Megatriorchis was sister to the clade of harriers (Circus) and that larger clade was sister to the Eurasian Sparrowhawk (Accipiter nisus).

Figure 1.

Maximum likelihood (single partition) phylogenetic tree for RAG-1 sequences for 68 species of accipitrids. Tree was rooted with four outgroups (not shown); branch lengths are drawn to scale. Positions of eight species in six genera endemic to the Australo-Papuan region are shown in red; position of a three basepair deletion delimiting four species of pernine kites is indicated by vertical blue bar. Likelihood bootstrap ranges are indicated for nodes with support of 50% or greater (open dots) and 80% or greater (solid dots). Some data used to infer this tree were originally presented in Griffiths et al. (2007).

The statistical bootstrap and Bayesian posterior support values for the positions of the six enigmatic genera, in the various phylogenetic analyses, are provided in Table 1. Support for a Hamirostra, Henicopernis, Lophoictinia clade within pernines was uniformly greater than 90%. The position of Harpyopsis with the harpy eagles and Macheiramphus, while present in all the shortest and most likely trees, had little bootstrap support, but a Bayesian posterior probability of 0.94. The relationships of Megatriorchis and Erythrotriorchis with accipiters and harriers had bootstrap support greater than 80% and a Bayesian posterior of 1.0.

Table 1.

Statistical support for important clades of hawks under alternate phylogenetic analyses.

The SOWH permutation likelihood ratio test provided robust evidence that the six genera of Australo-Papuan endemics were not members of a monophyletic clade. The difference between the log-likelihoods of the constrained and maximum likelihood trees, δ  =  192.84, exceeded the entire distribution of 100 randomized parametric log-likelihood scores by a multiple of more than nine; thus, P <<0.05 for a monophyletic clade of endemics.

Discussion